tag:blogger.com,1999:blog-28271018889963040292024-03-12T11:51:10.771-07:00T-RoboticsTerryhttp://www.blogger.com/profile/10286616328664292643noreply@blogger.comBlogger142125tag:blogger.com,1999:blog-2827101888996304029.post-71629099864303974672019-04-13T04:58:00.003-07:002019-04-13T04:58:44.549-07:00#44. KAIST에서 "대학 유죄"를 외치다: ART Lab의 도전<br />
3월 22일 카이스트 전산학부에서 진행된 AI x Education 포럼의 발표 내용입니다.<br />
대학은 과연 최적화된 교육을 제공하고 있을까요? 인공지능 기술을 배우려면 꼭 대학원에 가야 할까요?<br />
이 영상을 보시면 제가 요즘 어떤 교육을 꿈꾸고 어떤 일들을 벌이고 있는지 아실 수 있을 것입니다.<br />
<br />
인공지능/로보틱스 기술을 배우는 가장 쉬운 길, ART Lab 유튜브 채널의 구독, 좋아요 부탁드려요~!<br />
https://www.youtube.com/channel/UCzypbmDj_kVPDW3qWlrEFjA<br />
<div class="wp-caption aligncenter" id="attachment_25279" style="background-color: #f3f3f3; border-radius: 1px 3px 3px; border: 1px solid rgb(221, 221, 221); float: none; line-height: 21px; margin: 20px auto; max-width: 100%; outline: 0px; padding: 4px 0px 0px; text-align: center; vertical-align: baseline; width: 610px;">
<div style="color: #222222;">
<br />
<iframe allow="accelerometer; autoplay; encrypted-media; gyroscope; picture-in-picture" allowfullscreen="" frameborder="0" height="315" src="https://www.youtube.com/embed/DGNfe5359P4" width="560"></iframe></div>
<div style="color: #222222;">
<b><br /></b>
<br />
<div style="color: #222222;">
<iframe allow="accelerometer; autoplay; encrypted-media; gyroscope; picture-in-picture" allowfullscreen="" frameborder="0" height="315" src="https://www.youtube.com/embed/DF8b7futEDM" width="560"></iframe></div>
<div style="color: #222222;">
</div>
<b>[비디오]</b></div>
<div style="color: #222222;">
<br />
<div class="separator" style="clear: both; text-align: center;">
<iframe allowfullscreen="" frameborder="0" height="485" marginheight="0" marginwidth="0" scrolling="no" src="//www.slideshare.net/slideshow/embed_code/key/GivDHzn31pfgNf" style="border-width: 1px; border: 1px solid #ccc; margin-bottom: 5px; max-width: 100%;" width="595"> </iframe> </div>
<div style="margin-bottom: 5px;">
<strong> <a href="https://www.slideshare.net/TerryTaewoongUm/44-kaist-art-lab" target="_blank" title="#44. KAIST에서 "대학 유죄"를 외치다: ART Lab의 도전">#44. KAIST에서 "대학 유죄"를 외치다: ART Lab의 도전</a> </strong> from <strong><a href="https://www.slideshare.net/TerryTaewoongUm" target="_blank">Terry Taewoong Um</a></strong></div>
</div>
<div class="separator" style="clear: both; text-align: center;">
<b>[슬라이드]</b></div>
<div class="separator" style="clear: both; margin-bottom: 0px; margin-left: 0px; margin-right: 0px; margin-top: 0px; text-align: center;">
<b><br /></b></div>
<div style="font-family: "맑은 고딕"; margin: 0px;">
<div>
</div>
</div>
</div>
<span style="font-size: x-small;">* 테리의 딥러닝 토크 </span><br />
<span style="font-size: x-small;">[youtube] <a href="https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq%C2%A0" target="_blank">https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq </a></span><br />
<span style="font-size: x-small;">[facebook] <a href="https://www.facebook.com/deeplearningtalk/%C2%A0" target="_blank">https://www.facebook.com/deeplearningtalk/ </a></span><br />
<span style="font-size: x-small;"><br /></span><span style="font-size: x-small;">* T-Robotics의 글은 facebook과 rss reader로도 받아보실 수 있습니다. </span><br />
<span style="font-size: x-small;">[facebook] <a href="http://facebook.com/trobotics">http://facebook.com/trobotics</a> </span><br />
<span style="font-size: x-small;">[rss] <a href="http://t-robotics.blogspot.kr/feeds/posts/default">http://t-robotics.blogspot.kr/feeds/posts/default</a></span>
Terryhttp://www.blogger.com/profile/10286616328664292643noreply@blogger.com2tag:blogger.com,1999:blog-2827101888996304029.post-59850126509842070892019-03-07T06:42:00.001-08:002019-03-07T06:42:05.542-08:00#43. "정확도 98%의 인공지능"이 거짓인 이유<br />
작년 저는 KIST에 계신 선배님으로부터 자문을 요청 받았습니다. 학생이 너무 좋은 결과를 가지고 왔는데, 분명 뭐가 잘못된 것 같은데 한번 봐달라는 것이었습니다. 학생은 뇌파를 가지고 환자가 어떤 운동상태인지를 머신러닝으로 판단하고 있었는데, 정확도가 무려 95%가 넘었습니다.<br />
<br />
저는 자신있게 얘기했죠. 이거 잘못된 거 맞다고... 도대체 뭐가 잘못 되었을까요?레벨 별로 저지르는 실수를 정리해 보았습니다. 결론부터 말씀 드리자면, 현실 성능에 가깝게 성능을 보여주는 test set을 갖는 것이 곧 그 연구실/기업의 실력입니다. 너무 결과가 좋게 나오면 꼭 나를 의심합시다ㅎㅎ<br />
<div class="wp-caption aligncenter" id="attachment_25279" style="background-color: #f3f3f3; border-radius: 1px 3px 3px; border: 1px solid rgb(221, 221, 221); float: none; line-height: 21px; margin: 20px auto; max-width: 100%; outline: 0px; padding: 4px 0px 0px; text-align: center; vertical-align: baseline; width: 610px;">
<div style="color: #222222;">
<br />
<iframe allow="accelerometer; autoplay; encrypted-media; gyroscope; picture-in-picture" allowfullscreen="" frameborder="0" height="315" src="https://www.youtube.com/embed/neAJ9eldUWM" width="560"></iframe>
</div>
<div style="color: #222222;">
<b>[비디오]</b></div>
<div style="color: #222222;">
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEiCkVJAtoHzcDcsOOQjXIQVf6H2KKSm43C0nZJo4B9fEbwiN7pMyqBLYLm0ZibVw-hjkXWcXcOLjw4fnVA74e59sjimwBBi3-dtIDevm_RO1osKVGMykQkrXNVLmFZRMWr6lrGc_sOdfJmm/s1600/Slide1.PNG" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" data-original-height="720" data-original-width="960" height="300" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEiCkVJAtoHzcDcsOOQjXIQVf6H2KKSm43C0nZJo4B9fEbwiN7pMyqBLYLm0ZibVw-hjkXWcXcOLjw4fnVA74e59sjimwBBi3-dtIDevm_RO1osKVGMykQkrXNVLmFZRMWr6lrGc_sOdfJmm/s400/Slide1.PNG" width="400" /></a></div>
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEiPo5lshj-3K4gKrzKc4Mesf4m2_TveSBmBTUVsQAQNxd-pYT6LN-VDC2eM05pQvuMkLa0-ANVrYlkKsC1NH_aDrrem91rt0OzZrZ4eTz4tLrwDiUM2Syrc-_vS8G2cBGwqzmZfdQSQwNiu/s1600/Slide2.PNG" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" data-original-height="720" data-original-width="960" height="300" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEiPo5lshj-3K4gKrzKc4Mesf4m2_TveSBmBTUVsQAQNxd-pYT6LN-VDC2eM05pQvuMkLa0-ANVrYlkKsC1NH_aDrrem91rt0OzZrZ4eTz4tLrwDiUM2Syrc-_vS8G2cBGwqzmZfdQSQwNiu/s400/Slide2.PNG" width="400" /></a></div>
</div>
<div class="separator" style="clear: both; text-align: center;">
<b>[슬라이드]</b></div>
<div class="separator" style="clear: both; margin-bottom: 0px; margin-left: 0px; margin-right: 0px; margin-top: 0px; text-align: center;">
<b><br /></b></div>
<div style="font-family: "맑은 고딕"; margin: 0px;">
<div>
</div>
</div>
</div>
<span style="font-size: x-small;">* 테리의 딥러닝 토크 </span><br />
<span style="font-size: x-small;">[youtube] <a href="https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq%C2%A0" target="_blank">https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq </a></span><br />
<span style="font-size: x-small;">[facebook] <a href="https://www.facebook.com/deeplearningtalk/%C2%A0" target="_blank">https://www.facebook.com/deeplearningtalk/ </a></span><br />
<span style="font-size: x-small;"><br /></span><span style="font-size: x-small;">* T-Robotics의 글은 facebook과 rss reader로도 받아보실 수 있습니다. </span><br />
<span style="font-size: x-small;">[facebook] <a href="http://facebook.com/trobotics">http://facebook.com/trobotics</a> </span><br />
<span style="font-size: x-small;">[rss] <a href="http://t-robotics.blogspot.kr/feeds/posts/default">http://t-robotics.blogspot.kr/feeds/posts/default</a></span>
Terryhttp://www.blogger.com/profile/10286616328664292643noreply@blogger.com1tag:blogger.com,1999:blog-2827101888996304029.post-36318123607314817852019-02-13T22:23:00.000-08:002019-02-13T22:27:30.938-08:00#42. 사진 속 글자 읽기, OCR (Optical Character Recognition)<br />
카메라로 차량번호판을 인식하는 것은 이젠 너무나 당연한 기술이 되어 버렸죠?이렇게 사진 속에서 글자를 인식하는 기술을 OCR(optical character recognition)이라고 하는데요, 명함앱, 번역앱 등 정말 다양한 곳에서 쓰이는 기술이랍니다.<br />
<br />
혹시 이 문제가 이미 해결 되었다고 생각할지 모르겠지만, 아직도 OCR엔 과제들이 많이 남아있습니다. 그리고 최근엔 이러한 과제들을 딥러닝을 이용해 해결하려 하고있죠. OCR은 text localization과 recognition의 복합 기술인데요, 이를 위해 이미지 인식과 음성 인식의 요소 기술들이 복합적으로 쓰이죠.<br />
<br />
오늘은 OCR 리더보드 1등을 기록하셨던 이활석님의 DEVIEW 2018 발표자료를 바탕으로 OCR을 설명해보도록 하겠습니다.<br />
<div style="text-align: center;">
<div style="text-align: left;">
<br />
[Link] 이활석, "Reading AI" at DEVIEW2018, <a href="https://deview.kr/2018/schedule/235">https://deview.kr/2018/schedule/235</a><br />
<div>
<br /></div>
[YouTube] <span style="color: #0000ee;"><u><a href="https://www.youtube.com/watch?v=ckRFBl_XWFg&list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq&index=60">https://www.youtube.com/watch?v=ckRFBl_XWFg&list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq&index=60</a></u></span><br />
[Slides] <a href="https://www.slideshare.net/TerryTaewoongUm/a-brief-introduction-to-ocr-optical-character-recognition">https://www.slideshare.net/TerryTaewoongUm/a-brief-introduction-to-ocr-optical-character-recognition</a></div>
</div>
<div style="text-align: center;">
<div style="text-align: left;">
<br /></div>
</div>
<div class="wp-caption aligncenter" id="attachment_25279" style="background-color: #f3f3f3; border-radius: 1px 3px 3px; border: 1px solid rgb(221, 221, 221); float: none; line-height: 21px; margin: 20px auto; max-width: 100%; outline: 0px; padding: 4px 0px 0px; text-align: center; vertical-align: baseline; width: 610px;">
<div style="text-align: start;">
<br /></div>
<div style="color: #222222;">
<iframe allow="accelerometer; autoplay; encrypted-media; gyroscope; picture-in-picture" allowfullscreen="" frameborder="0" height="315" src="https://www.youtube.com/embed/ckRFBl_XWFg" width="560"></iframe>
</div>
<div style="color: #222222;">
<b>[비디오]</b></div>
<div style="color: #222222;">
<br /></div>
<div class="separator" style="clear: both; color: #222222;">
<br /></div>
<div class="separator" style="clear: both; color: #222222;">
<iframe allowfullscreen="" frameborder="0" height="485" marginheight="0" marginwidth="0" scrolling="no" src="//www.slideshare.net/slideshow/embed_code/key/a1TIA5d2BruXxu" style="border-width: 1px; border: 1px solid #ccc; margin-bottom: 5px; max-width: 100%;" width="595"> </iframe></div>
<div style="font-family: "맑은 고딕"; margin: 0px;">
<div>
</div>
</div>
[슬라이드]<br />
<div style="text-align: left;">
</div>
</div>
<br />
<span style="font-size: x-small;">* 테리의 딥러닝 토크 </span><br />
<span style="font-size: x-small;">[youtube] <a href="https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq%C2%A0" target="_blank">https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq </a></span><br />
<span style="font-size: x-small;">[facebook] <a href="https://www.facebook.com/deeplearningtalk/%C2%A0" target="_blank">https://www.facebook.com/deeplearningtalk/ </a></span><br />
<span style="font-size: x-small;"><br /></span><span style="font-size: x-small;">* T-Robotics의 글은 facebook과 rss reader로도 받아보실 수 있습니다. </span><br />
<span style="font-size: x-small;">[facebook] <a href="http://facebook.com/trobotics">http://facebook.com/trobotics</a> </span><br />
<span style="font-size: x-small;">[rss] <a href="http://t-robotics.blogspot.kr/feeds/posts/default">http://t-robotics.blogspot.kr/feeds/posts/default</a></span>
Terryhttp://www.blogger.com/profile/10286616328664292643noreply@blogger.com0tag:blogger.com,1999:blog-2827101888996304029.post-53010174369549848802019-01-28T03:59:00.002-08:002019-01-28T03:59:31.528-08:00#40. 딥러닝을 위한 PC 사양 2019<br />
안녕하세요. 괜히 공부가 안되니 '장비탓인가...' 싶은 생각이 드는 겨울입니다. 이럴 땐 질러야죠 ㅎㅎ<br />
<br />
테리의 딥러닝토크에서 "딥러닝을 위한 하드웨어 구성"에 대해 이야기를 나눈 게 2년 전쯤이었는데요, 업데이트 할 필요가 느껴 (그리고 저도 컴퓨터 한 대를 사고 싶어서ㅠ) 다시 한번 정리하였습니다.<br />
<br />
공부가 안되는 건 다 장비가 안좋아서 그런 겁니다. 새학기는 원래 학용품 새로 사고, 컴퓨터 새로 사야하는 시즌 아니겠습니까 ㅎㅎ 함께 질러보시죠 ㅋㅋ<br />
<br />
[PC 사양 정리]<br />
GPU : Nvidia RTX2070 8G (61만)<br />
CPU : AMD Ryzen7 2700x (31만)<br />
Mem : 삼성 PC4-21300 16G (12만)<br />
SSD : 삼성 860 EVO 500G (11만)<br />
HDD : 시게이트 2TB (7만)<br />
메인보드 : ASRock x470 Taichi 에즈원 (28만)<br />
PSU : FSP Hyper K 700W (7만)<br />
기타 : CPU쿨러, VGA쿨러, 쿨링 케이스 (11만)<br />
========<br />
총 168만원<br />
<div class="wp-caption aligncenter" id="attachment_25279" style="background-color: #f3f3f3; border-radius: 1px 3px 3px; border: 1px solid rgb(221, 221, 221); float: none; line-height: 21px; margin: 20px auto; max-width: 100%; outline: 0px; padding: 4px 0px 0px; text-align: center; vertical-align: baseline; width: 610px;">
<div style="color: #222222;">
<br />
<iframe allow="accelerometer; autoplay; encrypted-media; gyroscope; picture-in-picture" allowfullscreen="" frameborder="0" height="315" src="https://www.youtube.com/embed/9AgqC_NW3as" width="560"></iframe>
</div>
<div style="color: #222222;">
<b>[비디오]</b></div>
<div style="color: #222222;">
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEj51Jndhx1qS3cL4aSFCdusd3gUwYgC6hyphenhyphen5Yig-eR73H0IQJ87JrK3aMk1ZCYXyFbBE6vaI7E8c9nE0B02AFk5BDYGES1TZHLch2C5Jmh8_zKQhN7l0kXotYQ2ps8ZVlJJVzNiJE9HGdP8k/s1600/Slide1.PNG" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" data-original-height="720" data-original-width="960" height="300" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEj51Jndhx1qS3cL4aSFCdusd3gUwYgC6hyphenhyphen5Yig-eR73H0IQJ87JrK3aMk1ZCYXyFbBE6vaI7E8c9nE0B02AFk5BDYGES1TZHLch2C5Jmh8_zKQhN7l0kXotYQ2ps8ZVlJJVzNiJE9HGdP8k/s400/Slide1.PNG" width="400" /></a></div>
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEjuKpEM0UOeBetw-GSlgXHeP6LduAimfGoFJr8SpAfsgOSJKgpzL3pbScsr8PVSX5MH4dhk0cRmv3Cv59Rc57B0xHUSydkFXbGjPPB575_u0uGfUnIn7F4lgEGCmNoptqt7L1aonjhin-kb/s1600/Slide2.PNG" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" data-original-height="720" data-original-width="960" height="300" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEjuKpEM0UOeBetw-GSlgXHeP6LduAimfGoFJr8SpAfsgOSJKgpzL3pbScsr8PVSX5MH4dhk0cRmv3Cv59Rc57B0xHUSydkFXbGjPPB575_u0uGfUnIn7F4lgEGCmNoptqt7L1aonjhin-kb/s400/Slide2.PNG" width="400" /></a></div>
</div>
<div class="separator" style="clear: both; text-align: center;">
<b>[슬라이드]</b></div>
<div class="separator" style="clear: both; margin-bottom: 0px; margin-left: 0px; margin-right: 0px; margin-top: 0px; text-align: center;">
<b><br /></b></div>
<div style="font-family: "맑은 고딕"; margin: 0px;">
<div>
</div>
</div>
</div>
<span style="font-size: x-small;">* 테리의 딥러닝 토크 </span><br />
<span style="font-size: x-small;">[youtube] <a href="https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq%C2%A0" target="_blank">https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq </a></span><br />
<span style="font-size: x-small;">[facebook] <a href="https://www.facebook.com/deeplearningtalk/%C2%A0" target="_blank">https://www.facebook.com/deeplearningtalk/ </a></span><br />
<span style="font-size: x-small;"><br /></span><span style="font-size: x-small;">* T-Robotics의 글은 facebook과 rss reader로도 받아보실 수 있습니다. </span><br />
<span style="font-size: x-small;">[facebook] <a href="http://facebook.com/trobotics">http://facebook.com/trobotics</a> </span><br />
<span style="font-size: x-small;">[rss] <a href="http://t-robotics.blogspot.kr/feeds/posts/default">http://t-robotics.blogspot.kr/feeds/posts/default</a></span>
Terryhttp://www.blogger.com/profile/10286616328664292643noreply@blogger.com1tag:blogger.com,1999:blog-2827101888996304029.post-90532139915845449402018-12-01T16:24:00.000-08:002018-12-01T16:24:07.810-08:00#39. 딥러닝 공부 가이드 2019 (무료 강의, 책)테리가 추천하는 무료 딥러닝 공부자료들입니다.<br />
<br />
딥러닝 토크 처음 시작할 때도 한번 추천한 바 있는데, 그게 벌써 2년 전이더라고요. 그 때 추천해드린 자료도 여전히 좋은 자료들인데요, 새로운 자료들도 좋은게 많아 2년 만에 2019년 버전으로 업데이트 하였습니다!<br />
<br />
영상에서 소개하고 있는 무료 강좌 / 책 링크들은 다음과 같습니다.<br />
(유튜브 위주로 올렸으나 사이트들도 있는 경우가 많으니 참고하세요)<br />
<br />
[무료 강좌 - 섬머스쿨]<br />
머신러닝 1시간으로 입문하기 (Terry) https://goo.gl/FoPe2c<br />
Deep Learning & Reinforcement Learning Summer School<br />
(2017) https://goo.gl/4WthXN<br />
(2018) https://goo.gl/Z7stFe<br />
Machine Learning Summer School (MPI) https://goo.gl/5hQtA1<br />
<br />
[무료 강좌 - 한학기 강좌]<br />
DeepLearning.ai (Andrew Ng) https://goo.gl/mMr4AW<br />
CS231n (Stanford) https://goo.gl/WTLZkg<br />
CS224d (Stanford) https://goo.gl/nmY6Ws<br />
Fast.ai (Fast.ai) https://goo.gl/aBkesx<br />
<br />
[무료 강좌 - 강화학습]<br />
Deep RL bootcamp (UC Berkeley) https://goo.gl/i6CbtR<br />
UCL lectures (DeepMind) https://goo.gl/gF7EoY<br />
CS294 (UC Berkeley) https://goo.gl/d17a5x<br />
<br />
[무료 강좌 - 한글]<br />
모두의 딥러닝 (김성훈) https://goo.gl/zP6cYS<br />
모두의 강화학습 (김성훈) https://goo.gl/XQiB9f<br />
자연어 처리 (조경현) https://goo.gl/h8q6Kw<br />
딥러닝 논문들 (최성준) https://goo.gl/7f2YyN<br />
머신러닝 & 파이썬 (최성철, 유료) https://goo.gl/gy3U2y<br />
<br />
[무료 책]<br />
"Pattern Recognition and Machine Learning" https://goo.gl/EMbNKm<br />
"The Element of Statistical Learning" https://goo.gl/Y8GqqG<br />
"Gaussian Process for Machine Learning" https://goo.gl/4LU3Df<br />
"Dive into Deep Learning" https://goo.gl/Bk5wF5<br />
"Deep Learning" https://goo.gl/4kVPrm<br />
"Machine Learning: a Probabilistic Perspective" (유료)<br />
<div class="wp-caption aligncenter" id="attachment_25279" style="background-color: #f3f3f3; border-radius: 1px 3px 3px; border: 1px solid rgb(221, 221, 221); float: none; line-height: 21px; margin: 20px auto; max-width: 100%; outline: 0px; padding: 4px 0px 0px; text-align: center; vertical-align: baseline; width: 610px;">
<div style="color: #222222;">
<br />
<iframe allow="accelerometer; autoplay; encrypted-media; gyroscope; picture-in-picture" allowfullscreen="" frameborder="0" height="315" src="https://www.youtube.com/embed/fEwVeyki5p8" width="560"></iframe>
</div>
<div style="color: #222222;">
<b>[비디오]</b></div>
<div style="color: #222222;">
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEjfe07Sxz6qU8upywaL4dOrCaIv4OKyvbh93p9n6lN2Ts1maKiDAFfSczYPV2YMaQ3P_9dnXcDL91jYhkka0rSR9SUDKaKWDyMypq1x7VK-mtny5459mFoSeqvuNlOn5WycHqnRQs1pZBE-/s1600/%25EC%258A%25AC%25EB%259D%25BC%25EC%259D%25B4%25EB%2593%259C1.PNG" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" data-original-height="720" data-original-width="960" height="300" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEjfe07Sxz6qU8upywaL4dOrCaIv4OKyvbh93p9n6lN2Ts1maKiDAFfSczYPV2YMaQ3P_9dnXcDL91jYhkka0rSR9SUDKaKWDyMypq1x7VK-mtny5459mFoSeqvuNlOn5WycHqnRQs1pZBE-/s400/%25EC%258A%25AC%25EB%259D%25BC%25EC%259D%25B4%25EB%2593%259C1.PNG" width="400" /></a></div>
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEiLhnf28zOLhINeg9W3IiEnZl9_HzZPdWum6-CLRrjf1ee5YCBywE9ZsLHq4gFikYXqRvci_wb-IGD36Ne_4tHICFt-qgwcz5Vl54zn2Qp7Awbc3UdZ7eXsf2LAz43UkJdnLAA_UOXzGjJh/s1600/%25EC%258A%25AC%25EB%259D%25BC%25EC%259D%25B4%25EB%2593%259C2.PNG" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" data-original-height="720" data-original-width="960" height="300" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEiLhnf28zOLhINeg9W3IiEnZl9_HzZPdWum6-CLRrjf1ee5YCBywE9ZsLHq4gFikYXqRvci_wb-IGD36Ne_4tHICFt-qgwcz5Vl54zn2Qp7Awbc3UdZ7eXsf2LAz43UkJdnLAA_UOXzGjJh/s400/%25EC%258A%25AC%25EB%259D%25BC%25EC%259D%25B4%25EB%2593%259C2.PNG" width="400" /></a></div>
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEgcJOUMVaNHNssVQFXit0ZSZ21ggq485TwzADSpqdDvlNPCU0V8jtSQg3j_YjuTipiSkhd-I38twkntnBSy4ab2IOOmeR5BWoHyMZvOmYvwnDJLuOq0VbE3eDSmJkyidaHNTg1AzpI8TMZy/s1600/%25EC%258A%25AC%25EB%259D%25BC%25EC%259D%25B4%25EB%2593%259C3.PNG" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" data-original-height="720" data-original-width="960" height="300" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEgcJOUMVaNHNssVQFXit0ZSZ21ggq485TwzADSpqdDvlNPCU0V8jtSQg3j_YjuTipiSkhd-I38twkntnBSy4ab2IOOmeR5BWoHyMZvOmYvwnDJLuOq0VbE3eDSmJkyidaHNTg1AzpI8TMZy/s400/%25EC%258A%25AC%25EB%259D%25BC%25EC%259D%25B4%25EB%2593%259C3.PNG" width="400" /></a></div>
</div>
<div class="separator" style="clear: both; text-align: center;">
<b>[슬라이드]</b></div>
<div class="separator" style="clear: both; margin-bottom: 0px; margin-left: 0px; margin-right: 0px; margin-top: 0px; text-align: center;">
<b><br /></b></div>
<div style="font-family: "맑은 고딕"; margin: 0px;">
<div>
</div>
</div>
</div>
<span style="font-size: x-small;">* 테리의 딥러닝 토크 </span><br />
<span style="font-size: x-small;">[youtube] <a href="https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq%C2%A0" target="_blank">https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq </a></span><br />
<span style="font-size: x-small;">[facebook] <a href="https://www.facebook.com/deeplearningtalk/%C2%A0" target="_blank">https://www.facebook.com/deeplearningtalk/ </a></span><br />
<span style="font-size: x-small;"><br /></span><span style="font-size: x-small;">* T-Robotics의 글은 facebook과 rss reader로도 받아보실 수 있습니다. </span><br />
<span style="font-size: x-small;">[facebook] <a href="http://facebook.com/trobotics">http://facebook.com/trobotics</a> </span><br />
<span style="font-size: x-small;">[rss] <a href="http://t-robotics.blogspot.kr/feeds/posts/default">http://t-robotics.blogspot.kr/feeds/posts/default</a></span>
Terryhttp://www.blogger.com/profile/10286616328664292643noreply@blogger.com0tag:blogger.com,1999:blog-2827101888996304029.post-46840405530694715962018-11-08T21:31:00.004-08:002018-11-08T21:31:33.357-08:00#38. 딥러닝의 실제문제 적용, 레이블이 전부다연구든 비지니스든, 실제 데이터에 딥러닝을 적용하시는 분들이라면 아마 supervised learning을 먼저 고려하셔야 할텐데요, 최대한 데이터를 빨리 모은 후 여러 최신 딥러닝 모델들을 적용하고 싶으시겠죠.<br />
<br />
그런데 저는 다시한번 데이터 그 자체에 집중하셔야 한다고 말씀드리고 싶습니다. 왜냐면 좋은 데이터를 이기는 머신러닝 모델은 없거든요.<br />
<br />
이 영상의 주제는 다음과 같습니다.<br />
<br />
- 좋은 레이블을 이기는 머신러닝 모델은 없다.<br />
- supervised learning은 invariance를 배우는 과정이다.<br />
- 말로 표현할 수 있다면 최대한 그것을 적용하는 것이 데이터 효율을 높인다.<br />
- 좋은 domain-knolwedge를 통해 annotation 알고리즘을 만들고, 그것을 통해 labeled dataset을 얻을텐데, 그래도 데이터는 한땀한땀 점검을 해야한다.<br />
<br />
그럼 즐겁게 들어주세요 ^^<br />
<div class="wp-caption aligncenter" id="attachment_25279" style="background-color: #f3f3f3; border-radius: 1px 3px 3px; border: 1px solid rgb(221, 221, 221); float: none; line-height: 21px; margin: 20px auto; max-width: 100%; outline: 0px; padding: 4px 0px 0px; text-align: center; vertical-align: baseline; width: 610px;">
<div style="color: #222222;">
<br />
<iframe allow="accelerometer; autoplay; encrypted-media; gyroscope; picture-in-picture" allowfullscreen="" frameborder="0" height="315" src="https://www.youtube.com/embed/DuTROXxnSdw" width="560"></iframe>
</div>
<div style="color: #222222;">
<b>[비디오]</b></div>
<div style="color: #222222;">
<br />
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEj4i3uVbNXmAFmXmSjVq81NlzpjZb0-FMglyGNXm9C0nHSEH8CH4GYLgjaPPczlOxFvhS_HpziHRZWgPx9PVIKe8DnFmC8n5NUecDqI0Jbgl6lIvOECv8Ltzx2iEjf6V7KfpnOVBX4SmyEi/s1600/%25EC%258A%25AC%25EB%259D%25BC%25EC%259D%25B4%25EB%2593%259C1.PNG" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" data-original-height="720" data-original-width="960" height="300" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEj4i3uVbNXmAFmXmSjVq81NlzpjZb0-FMglyGNXm9C0nHSEH8CH4GYLgjaPPczlOxFvhS_HpziHRZWgPx9PVIKe8DnFmC8n5NUecDqI0Jbgl6lIvOECv8Ltzx2iEjf6V7KfpnOVBX4SmyEi/s400/%25EC%258A%25AC%25EB%259D%25BC%25EC%259D%25B4%25EB%2593%259C1.PNG" width="400" /></a><br />
<br /></div>
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEhTnHtp6sk4BZVOuaR4Odwsci_kadE6Tg0poso0RIBiyVKXRRu9mPDTS2CANaPJjfO4XPlH33zRGi1NZ4n7PZjfSaYO45ml0O-Qd8EvG0NqJu0LsE9vDudFXx6TP0adP9p8Qct5n0gmDi9k/s1600/%25EC%258A%25AC%25EB%259D%25BC%25EC%259D%25B4%25EB%2593%259C2.PNG" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" data-original-height="720" data-original-width="960" height="300" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEhTnHtp6sk4BZVOuaR4Odwsci_kadE6Tg0poso0RIBiyVKXRRu9mPDTS2CANaPJjfO4XPlH33zRGi1NZ4n7PZjfSaYO45ml0O-Qd8EvG0NqJu0LsE9vDudFXx6TP0adP9p8Qct5n0gmDi9k/s400/%25EC%258A%25AC%25EB%259D%25BC%25EC%259D%25B4%25EB%2593%259C2.PNG" width="400" /></a></div>
<div class="separator" style="clear: both; text-align: center;">
<b>[슬라이드]</b></div>
<div class="separator" style="clear: both; margin-bottom: 0px; margin-left: 0px; margin-right: 0px; margin-top: 0px; text-align: center;">
<b><br /></b></div>
<div style="font-family: "맑은 고딕"; margin: 0px;">
<div>
</div>
</div>
</div>
<span style="font-size: x-small;">* 테리의 딥러닝 토크 </span><br />
<span style="font-size: x-small;">[youtube] <a href="https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq%C2%A0" target="_blank">https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq </a></span><br />
<span style="font-size: x-small;">[facebook] <a href="https://www.facebook.com/deeplearningtalk/%C2%A0" target="_blank">https://www.facebook.com/deeplearningtalk/ </a></span><br />
<span style="font-size: x-small;"><br /></span><span style="font-size: x-small;">* T-Robotics의 글은 facebook과 rss reader로도 받아보실 수 있습니다. </span><br />
<span style="font-size: x-small;">[facebook] <a href="http://facebook.com/trobotics">http://facebook.com/trobotics</a> </span><br />
<span style="font-size: x-small;">[rss] <a href="http://t-robotics.blogspot.kr/feeds/posts/default">http://t-robotics.blogspot.kr/feeds/posts/default</a></span>
Terryhttp://www.blogger.com/profile/10286616328664292643noreply@blogger.com0tag:blogger.com,1999:blog-2827101888996304029.post-17124804949164378982018-09-29T06:50:00.000-07:002018-09-29T06:50:08.282-07:00# 37. GIST 레이드랩에서 진행한 5분 잡썰"저 딥러닝 강좌도 다 봤고 시중에 나온 책도 다 봤는데 그 다음엔 뭘 해야하죠? 뭔가 아는 것 같으면서도 딥러닝을 잘 모르고 있는 것 같아요."<br />
<br />
많은 분들이 딥러닝 튜토리얼 레벨의 학습을 마친 후 무엇을 공부해야 하는지 질문을 하십니다. 아마도 다음 타겟은 튜토리얼 속 문제가 아닌 real-world problem에 도전하는 것이겠죠?<br />
<br />
하지만 real-world problem을 다루는 머신러닝 연구자/개발자는 튜토리얼에선 볼 수 없었던 많은 문제에 부딪치게 됩니다. 어쩌면 '뭐야.. 내가 현재 하고있는 일의 90%는 튜토리얼에서 중요하다고 언급도 안했던 내용들이잖아'라며 느끼실지도 모릅니다.<br />
<br />
과연 튜토리얼 밖 실제문제에 딥러닝을 적용할 땐 어떤 이슈들을 맞이하게 될까요? 오랜만에 돌아온 테리의 딥러닝톡, 즐겁게 들어주세요 :)<br />
<div class="wp-caption aligncenter" id="attachment_25279" style="background-color: #f3f3f3; border-radius: 1px 3px 3px; border: 1px solid rgb(221, 221, 221); float: none; line-height: 21px; margin: 20px auto; max-width: 100%; outline: 0px; padding: 4px 0px 0px; text-align: center; vertical-align: baseline; width: 610px;">
<div style="color: #222222;">
<br />
<iframe allowfullscreen="" frameborder="0" height="252" src="https://www.youtube.com/embed/daTqajvdeTw?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq" width="448"></iframe><br /></div>
<div style="color: #222222;">
<b>[비디오]</b></div>
<div style="color: #222222;">
<br /></div>
<div class="separator" style="clear: both; color: #222222;">
<br /></div>
<div class="separator" style="clear: both; color: #222222;">
</div>
<div class="separator" style="clear: both;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEjTx0nNFXr6aonu6jQoJheqOv1aFvijx87hQ6a0yh3vzytqMYt-o-GKqjoAduBhoGuetQGV7nzfypG33fslnbDKDRObDaRYSl4_38nf02QBlk1mr0JwrE-2vBuOn1ZLyw4wLkxdTjDO46KD/s1600/%25EC%258A%25AC%25EB%259D%25BC%25EC%259D%25B4%25EB%2593%259C1.PNG" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" data-original-height="720" data-original-width="960" height="300" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEjTx0nNFXr6aonu6jQoJheqOv1aFvijx87hQ6a0yh3vzytqMYt-o-GKqjoAduBhoGuetQGV7nzfypG33fslnbDKDRObDaRYSl4_38nf02QBlk1mr0JwrE-2vBuOn1ZLyw4wLkxdTjDO46KD/s400/%25EC%258A%25AC%25EB%259D%25BC%25EC%259D%25B4%25EB%2593%259C1.PNG" width="400" /></a></div>
<br />
<div class="separator" style="clear: both;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEiZwtc1nx9u2yMpyQ9l4KSU6XQcQ6IUaq6dpFYP9WlGS4AjKyml6AAgMucBtev7WFJN2LuXRb9qeJn-YBTJO6NBbgC345uHIKwJln7gKDtq8KibpEqcsPkMW4KxMIztd2UtcZ8-qfU03ISu/s1600/%25EC%258A%25AC%25EB%259D%25BC%25EC%259D%25B4%25EB%2593%259C2.PNG" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" data-original-height="720" data-original-width="960" height="300" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEiZwtc1nx9u2yMpyQ9l4KSU6XQcQ6IUaq6dpFYP9WlGS4AjKyml6AAgMucBtev7WFJN2LuXRb9qeJn-YBTJO6NBbgC345uHIKwJln7gKDtq8KibpEqcsPkMW4KxMIztd2UtcZ8-qfU03ISu/s400/%25EC%258A%25AC%25EB%259D%25BC%25EC%259D%25B4%25EB%2593%259C2.PNG" width="400" /></a></div>
<div class="separator" style="clear: both;">
<b>[슬라이드]</b></div>
<div class="separator" style="clear: both;">
<b><br /></b></div>
<div style="font-family: "맑은 고딕"; margin: 0px;">
<div>
</div>
</div>
</div>
<span style="font-size: x-small;">* 테리의 딥러닝 토크 </span><br />
<span style="font-size: x-small;">[youtube] <a href="https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq%C2%A0" target="_blank">https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq </a></span><br />
<span style="font-size: x-small;">[facebook] <a href="https://www.facebook.com/deeplearningtalk/%C2%A0" target="_blank">https://www.facebook.com/deeplearningtalk/ </a></span><br />
<span style="font-size: x-small;"><br /></span><span style="font-size: x-small;">* T-Robotics의 글은 facebook과 rss reader로도 받아보실 수 있습니다. </span><br />
<span style="font-size: x-small;">[facebook] <a href="http://facebook.com/trobotics">http://facebook.com/trobotics</a> </span><br />
<span style="font-size: x-small;">[rss] <a href="http://t-robotics.blogspot.kr/feeds/posts/default">http://t-robotics.blogspot.kr/feeds/posts/default</a></span>Terryhttp://www.blogger.com/profile/10286616328664292643noreply@blogger.com0tag:blogger.com,1999:blog-2827101888996304029.post-60676169070318477852018-09-20T18:57:00.000-07:002018-09-29T06:46:04.326-07:00#36. 딥러닝의 실제적용, 튜토리얼과 무엇이 다른가?"저 딥러닝 강좌도 다 봤고 시중에 나온 책도 다 봤는데 그 다음엔 뭘 해야하죠? 뭔가 아는 것 같으면서도 딥러닝을 잘 모르고 있는 것 같아요."<br />
<br />
많은 분들이 딥러닝 튜토리얼 레벨의 학습을 마친 후 무엇을 공부해야 하는지 질문을 하십니다. 아마도 다음 타겟은 튜토리얼 속 문제가 아닌 real-world problem에 도전하는 것이겠죠?<br />
<br />
하지만 real-world problem을 다루는 머신러닝 연구자/개발자는 튜토리얼에선 볼 수 없었던 많은 문제에 부딪치게 됩니다. 어쩌면 '뭐야.. 내가 현재 하고있는 일의 90%는 튜토리얼에서 중요하다고 언급도 안했던 내용들이잖아'라며 느끼실지도 모릅니다.<br />
<br />
과연 튜토리얼 밖 실제문제에 딥러닝을 적용할 땐 어떤 이슈들을 맞이하게 될까요? 오랜만에 돌아온 테리의 딥러닝톡, 즐겁게 들어주세요 :)<br />
<div class="wp-caption aligncenter" id="attachment_25279" style="background-color: #f3f3f3; border-radius: 1px 3px 3px; border: 1px solid rgb(221, 221, 221); float: none; line-height: 21px; margin: 20px auto; max-width: 100%; outline: 0px; padding: 4px 0px 0px; text-align: center; vertical-align: baseline; width: 610px;">
<div style="color: #222222;">
<br />
<iframe allowfullscreen="" frameborder="0" height="252" src="https://www.youtube.com/embed/daTqajvdeTw?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq" width="448"></iframe><br /></div>
<div style="color: #222222;">
<b>[비디오]</b></div>
<div style="color: #222222;">
<br /></div>
<div class="separator" style="clear: both; color: #222222;">
<br /></div>
<div class="separator" style="clear: both; color: #222222;">
</div>
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEjTx0nNFXr6aonu6jQoJheqOv1aFvijx87hQ6a0yh3vzytqMYt-o-GKqjoAduBhoGuetQGV7nzfypG33fslnbDKDRObDaRYSl4_38nf02QBlk1mr0JwrE-2vBuOn1ZLyw4wLkxdTjDO46KD/s1600/%25EC%258A%25AC%25EB%259D%25BC%25EC%259D%25B4%25EB%2593%259C1.PNG" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" data-original-height="720" data-original-width="960" height="300" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEjTx0nNFXr6aonu6jQoJheqOv1aFvijx87hQ6a0yh3vzytqMYt-o-GKqjoAduBhoGuetQGV7nzfypG33fslnbDKDRObDaRYSl4_38nf02QBlk1mr0JwrE-2vBuOn1ZLyw4wLkxdTjDO46KD/s400/%25EC%258A%25AC%25EB%259D%25BC%25EC%259D%25B4%25EB%2593%259C1.PNG" width="400" /></a></div>
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEiZwtc1nx9u2yMpyQ9l4KSU6XQcQ6IUaq6dpFYP9WlGS4AjKyml6AAgMucBtev7WFJN2LuXRb9qeJn-YBTJO6NBbgC345uHIKwJln7gKDtq8KibpEqcsPkMW4KxMIztd2UtcZ8-qfU03ISu/s1600/%25EC%258A%25AC%25EB%259D%25BC%25EC%259D%25B4%25EB%2593%259C2.PNG" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" data-original-height="720" data-original-width="960" height="300" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEiZwtc1nx9u2yMpyQ9l4KSU6XQcQ6IUaq6dpFYP9WlGS4AjKyml6AAgMucBtev7WFJN2LuXRb9qeJn-YBTJO6NBbgC345uHIKwJln7gKDtq8KibpEqcsPkMW4KxMIztd2UtcZ8-qfU03ISu/s400/%25EC%258A%25AC%25EB%259D%25BC%25EC%259D%25B4%25EB%2593%259C2.PNG" width="400" /></a></div>
<div class="separator" style="clear: both; margin-bottom: 0px; margin-left: 0px; margin-right: 0px; margin-top: 0px; text-align: center;">
<b>[슬라이드]</b></div>
<div class="separator" style="clear: both; margin-bottom: 0px; margin-left: 0px; margin-right: 0px; margin-top: 0px; text-align: center;">
<b><br /></b></div>
<div style="font-family: "맑은 고딕"; margin: 0px;">
<div>
</div>
</div>
</div>
<span style="font-size: x-small;">* 테리의 딥러닝 토크 </span><br />
<span style="font-size: x-small;">[youtube] <a href="https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq%C2%A0" target="_blank">https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq </a></span><br />
<span style="font-size: x-small;">[facebook] <a href="https://www.facebook.com/deeplearningtalk/%C2%A0" target="_blank">https://www.facebook.com/deeplearningtalk/ </a></span><br />
<span style="font-size: x-small;"><br /></span><span style="font-size: x-small;">* T-Robotics의 글은 facebook과 rss reader로도 받아보실 수 있습니다. </span><br />
<span style="font-size: x-small;">[facebook] <a href="http://facebook.com/trobotics">http://facebook.com/trobotics</a> </span><br />
<span style="font-size: x-small;">[rss] <a href="http://t-robotics.blogspot.kr/feeds/posts/default">http://t-robotics.blogspot.kr/feeds/posts/default</a></span>
Terryhttp://www.blogger.com/profile/10286616328664292643noreply@blogger.com0tag:blogger.com,1999:blog-2827101888996304029.post-58478113622826179302018-05-21T04:32:00.001-07:002018-09-29T06:45:29.922-07:00#35. Deep Variational Bayes Filters (2017)seqeunce나 time-series 데이터엔 RNN이나 LSTM이 참 효과적이죠? 하지만 그것만으로는 불만족스럽다고 하네요. "What's wrong with RNN?"<br />
<br />
사실 비슷한 질문을 한적이 있었죠. "Autoencoder가 있는데 Variational Autoencoder는 도대체 왜 있는거죠?" 핵심은 Latent space에 있습니다. 사실 우리는 prediction만을 내뱉는걸 원하지 않고, 확률분포를 알기를 바라며, 더 정확히는 observation을 만들어내는 데이터가 "진짜로" 살고 있는 공간, latent space를 찾길 바라는거죠.<br />
<br />
무슨 얘기냐고요? 이 발표를 들으시면 조금은 감이 잡히시지 않을까 싶습니다. VAE를 timeseries에 확장한, 혹은 칼만필터에 딥러닝을 첨가한 "Deep Variational Bayes Filters (2017)"이란 논문입니다.<br />
<br />
<br />
"Deep Variational Bayes Filters: Unsupervised Learning of State Space Models from Raw Data", ICLR2017.<br />
[Link] <a href="https://arxiv.org/abs/1605.06432">https://arxiv.org/abs/1605.06432</a><br />
<div style="text-align: center;">
<div style="text-align: left;">
<br /></div>
</div>
<div style="text-align: center;">
<div style="text-align: left;">
[YouTube] <a href="https://www.youtube.com/watch?v=uM0rQtL6_AA&t=61s&list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq&index=40">https://www.youtube.com/watch?v=uM0rQtL6_AA&t=61s&list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq&index=40</a><br />
[Slides] <a href="https://www.slideshare.net/TerryTaewoongUm/deep-variational-bayes-filters-2017">https://www.slideshare.net/TerryTaewoongUm/deep-variational-bayes-filters-2017</a></div>
</div>
<div style="text-align: center;">
<div style="text-align: left;">
<br /></div>
</div>
<div class="wp-caption aligncenter" id="attachment_25279" style="background-color: #f3f3f3; border-radius: 1px 3px 3px; border: 1px solid rgb(221, 221, 221); float: none; line-height: 21px; margin: 20px auto; max-width: 100%; outline: 0px; padding: 4px 0px 0px; text-align: center; vertical-align: baseline; width: 610px;">
<div style="text-align: start;">
<br /></div>
<div style="color: #222222;">
<iframe allowfullscreen="" frameborder="0" height="252" src="https://www.youtube.com/embed/uM0rQtL6_AA?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq" width="448"></iframe><br /></div>
<div style="color: #222222;">
<b>[비디오]</b></div>
<div style="color: #222222;">
<br /></div>
<div class="separator" style="clear: both; color: #222222;">
<br /></div>
<div class="separator" style="clear: both; color: #222222;">
<iframe allowfullscreen="" frameborder="0" height="485" marginheight="0" marginwidth="0" scrolling="no" src="https://www.slideshare.net/slideshow/embed_code/key/xlGF1viWzVy7Qb" style="border-color: rgb(204, 204, 204); border-style: solid; border-width: 1px; margin-bottom: 5px; max-width: 100%;" width="595"></iframe></div>
<div style="color: #222222;">
<b>[슬라이드]</b></div>
<div style="font-family: "맑은 고딕"; margin: 0px;">
<div>
</div>
</div>
</div>
<span style="font-size: x-small;">* 테리의 딥러닝 토크 </span><br />
<span style="font-size: x-small;">[youtube] <a href="https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq%C2%A0" target="_blank">https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq </a></span><br />
<span style="font-size: x-small;">[facebook] <a href="https://www.facebook.com/deeplearningtalk/%C2%A0" target="_blank">https://www.facebook.com/deeplearningtalk/ </a></span><br />
<span style="font-size: x-small;"><br /></span><span style="font-size: x-small;">* T-Robotics의 글은 facebook과 rss reader로도 받아보실 수 있습니다. </span><br />
<span style="font-size: x-small;">[facebook] <a href="http://facebook.com/trobotics">http://facebook.com/trobotics</a> </span><br />
<span style="font-size: x-small;">[rss] <a href="http://t-robotics.blogspot.kr/feeds/posts/default">http://t-robotics.blogspot.kr/feeds/posts/default</a></span>Terryhttp://www.blogger.com/profile/10286616328664292643noreply@blogger.com0tag:blogger.com,1999:blog-2827101888996304029.post-35935151738728401032018-04-13T06:20:00.001-07:002018-04-13T06:20:41.463-07:00#34. KL divergence, 순서가 중요할까요?Kullback-Leibler Divergence (쿨벡-라이블러 다이버전스, 혹은 KL divergence) 많이 써보셨죠? 두개의 확률분포가 있을 때 둘 사이의 차이를 측정해주는 용도로 많이 쓰는데요, 아마 'KL Divergence는 거리처럼 쓰이지만 거리는 아니다'라는 얘기도 많이 들어보셨을 겁니다.<br />
<br />
이유는 두 분포의 순서 중 어떤 것을 먼저 넣느냐에 따라 나오는 값이 달라지기 때문인데요, 즉 KL(P,Q) != KL(Q,P)라는거죠. 그렇다면... P(X)를 true, Q(X)를 predicted라고 했을 때 어떤 값을 먼저 넣어서 거리를 재야할까요? 순서에 따라 물리적 의미의 차이가 있을까요? 오늘은 KL-divergence에 대해 더 깊이 알아보았습니다.<br />
<br />
<br />
<div class="wp-caption aligncenter" id="attachment_25279" style="background-color: #f3f3f3; border-radius: 1px 3px 3px; border: 1px solid rgb(221, 221, 221); color: #222222; float: none; line-height: 21px; margin: 20px auto; max-width: 100%; outline: 0px; padding: 4px 0px 0px; text-align: center; vertical-align: baseline; width: 610px;">
<br />
<iframe allowfullscreen="" frameborder="0" height="252" src="https://www.youtube.com/embed/c5nTnvGHG4E?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq" width="448"></iframe><br />
<b>[비디오]</b><br />
<div class="separator" style="clear: both; text-align: center;">
<br /></div>
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEg3TgSDJjz3ibFiIfTQ3cKpgGQ3lLo9_rxgmFE3kHR_aLp1tHjpdtyxaqES05OWRwBPonLdjYTRlVvxSJLy50t_vvtKrdLTG7f7otV-ywZmidcnpD7FT8Eg7m5YWvtwBOFF7lRjkIw77_V9/s1600/Slide1.PNG" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" data-original-height="720" data-original-width="960" height="300" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEg3TgSDJjz3ibFiIfTQ3cKpgGQ3lLo9_rxgmFE3kHR_aLp1tHjpdtyxaqES05OWRwBPonLdjYTRlVvxSJLy50t_vvtKrdLTG7f7otV-ywZmidcnpD7FT8Eg7m5YWvtwBOFF7lRjkIw77_V9/s400/Slide1.PNG" width="400" /></a></div>
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEixZ8WEdGN7mU-dzgpjpYe8NNDNs5k5Be_eY45JRS_YtuMZ2g7SzQmUUavX6QqXZ9F1RqiYu5ZYbJLJfBAss7D0Kf7qnCg25Vj4eenQfu8_x6ULIHbcaeC5z__F2r2n4JLG6jou9bInMU6G/s1600/Slide2.PNG" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" data-original-height="720" data-original-width="960" height="300" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEixZ8WEdGN7mU-dzgpjpYe8NNDNs5k5Be_eY45JRS_YtuMZ2g7SzQmUUavX6QqXZ9F1RqiYu5ZYbJLJfBAss7D0Kf7qnCg25Vj4eenQfu8_x6ULIHbcaeC5z__F2r2n4JLG6jou9bInMU6G/s400/Slide2.PNG" width="400" /></a></div>
<div class="separator" style="clear: both;">
<b>[슬라이드]</b></div>
<span style="font-size: x-small;"><br /></span></div>
<span style="font-size: x-small;"><br /></span><span style="font-size: x-small;">* 테리의 딥러닝 토크 </span><br />
<span style="font-size: x-small;">[youtube] <a href="https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq">https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq</a> </span><br />
<span style="font-size: x-small;">[facebook] <a href="https://www.facebook.com/deeplearningtalk/">https://www.facebook.com/deeplearningtalk/</a> </span><br />
<span style="font-size: x-small;"><br /></span><span style="font-size: x-small;">* T-Robotics의 글은 facebook과 rss reader로도 받아보실 수 있습니다. </span><br />
<span style="font-size: x-small;">[facebook] <a href="http://facebook.com/trobotics">http://facebook.com/trobotics</a> </span><br />
<span style="font-family: "malgun gothic"; font-size: x-small;">[rss] </span><a href="http://t-robotics.blogspot.kr/feeds/posts/default" style="font-family: "malgun gothic"; font-size: small;">http://t-robotics.blogspot.kr/feeds/posts/default</a>Terryhttp://www.blogger.com/profile/10286616328664292643noreply@blogger.com1tag:blogger.com,1999:blog-2827101888996304029.post-81995659404308255652018-03-25T07:19:00.000-07:002018-03-25T07:19:34.453-07:00#33. On Calibration of Modern Neural Networks딥러닝으로 classification을 하면 "클래스 확률"이라 믿어지는 softmax 아웃풋이 나오죠? 하지만 아웃풋 노드값이 0.9가 나왔다고해서, 그것이 90% 확률로 맞을 가능성이 있다는 뜻은 아니랍니다. 그렇다면 0.9가 실제 "90% 정확도"를 표현하게 만들려면 어떻게 해야할까요?<br>
<br>
이러한 주제를 calibration이라고 하는데요, 이번에 PR-12 논문읽기 모임에서 발표한 "On Calibration of Modern Neural Networks"이란 논문은 단순히 상수로 나눠주는 것만으로 이 문제를 간단하게 해결하였습니다. 논문의 내용도 흥미롭지만, 문제를 설정하고 풀어나가는 과정도 연구자로서 볼만하네요. 그럼 즐겁게 들어주세요 ^^<br>
<a href="http://t-robotics.blogspot.com/2018/03/33-on-calibration-of-modern-neural.html#more">Read more »</a>Terryhttp://www.blogger.com/profile/10286616328664292643noreply@blogger.com0tag:blogger.com,1999:blog-2827101888996304029.post-15664075270043036432018-02-16T21:32:00.000-08:002018-02-16T21:32:07.190-08:00#32. PCA의 이해 PCA라고 들어보셨나요? 보통 차원축소(dimensionality reduction)나 시각화(visualization), 그리고 특징 추출(feature extraction)을 할 때 많이 쓰는 방법인데요, 사실 PCA는 모든 dimensionality reduction 기법들의 기본이자 선형대수의 종합일 정도로 너무나 중요하고 꼭 숙지해야 할 기법입니다. 아름다울 뿐만 아니라 실용적이기도 하고요.<br />
<br />
그래서 오늘은 PCA를 "수식없이" 살펴봤습니다. 내용이 중요하다보니 이야기가 좀 길어졌네요ㅎㅎ 영상을 보고 PCA에 대해 더 알아보실 분들은 저의 옛 블로그를 참고하시거나, 더 좋은 방법은 영어로 된 튜토리얼 문서를 한번 꼭 읽어보세요!<br />
<br />
[Link] "PCA의 의미와 한계점" <a href="https://www.facebook.com/TRobotics/posts/796663103771140">https://www.facebook.com/TRobotics/posts/796663103771140</a><br />
[Link] "A Tutorial on PCA", Jonathon Shlens, <a href="https://arxiv.org/abs/1404.1100">https://arxiv.org/abs/1404.1100</a><br />
<br />
<div class="wp-caption aligncenter" id="attachment_25279" style="background-color: #f3f3f3; border-radius: 1px 3px 3px; border: 1px solid rgb(221, 221, 221); color: #222222; float: none; line-height: 21px; margin: 20px auto; max-width: 100%; outline: 0px; padding: 4px 0px 0px; text-align: center; vertical-align: baseline; width: 610px;">
<br />
<iframe allowfullscreen="" frameborder="0" height="252" src="https://www.youtube.com/embed/9UggjVi9-9M?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq" width="448"></iframe><br />
<br />
<iframe allowfullscreen="" frameborder="0" height="252" src="https://www.youtube.com/embed/kMYRnyvtLC8?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq" width="448"></iframe><br />
<b>[비디오]</b><br />
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEjJvwNnUVgNNE-5S_h8sMVOY87xM1ntuFF1TJvyZKMbRMLCrhEgQLyzHAO-c5ej8XZueknO9ZNH6SynZbjsIg0tgyn7SqgrKPK2iJESvDDtk2vTwUUPEqM0LeOZM6i_xYnOxWkLzJB8CZmJ/s1600/Slide1.PNG" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" data-original-height="720" data-original-width="960" height="300" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEjJvwNnUVgNNE-5S_h8sMVOY87xM1ntuFF1TJvyZKMbRMLCrhEgQLyzHAO-c5ej8XZueknO9ZNH6SynZbjsIg0tgyn7SqgrKPK2iJESvDDtk2vTwUUPEqM0LeOZM6i_xYnOxWkLzJB8CZmJ/s400/Slide1.PNG" width="400" /></a></div>
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEgft4HqSY5EbmTWr5QoZUqrKYGKmy0BtAzhN4nB1tiCwdcqL4Fu8GcHvBhMmRqwEmpgEV4PXUjuYyrBQrwIuCOlsAcW4dDaY9t6vHDppMHrStTYiWIWDAcVmCv9OCH9b1o2a_2qpZ5Yc72A/s1600/Slide2.PNG" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" data-original-height="720" data-original-width="960" height="300" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEgft4HqSY5EbmTWr5QoZUqrKYGKmy0BtAzhN4nB1tiCwdcqL4Fu8GcHvBhMmRqwEmpgEV4PXUjuYyrBQrwIuCOlsAcW4dDaY9t6vHDppMHrStTYiWIWDAcVmCv9OCH9b1o2a_2qpZ5Yc72A/s400/Slide2.PNG" width="400" /></a></div>
<div class="separator" style="clear: both;">
<b>[슬라이드]</b></div>
<span style="font-size: x-small;"><br /></span></div>
<span style="font-size: x-small;"><br /></span><span style="font-size: x-small;">* 테리의 딥러닝 토크 </span><br />
<span style="font-size: x-small;">[youtube] <a href="https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq">https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq</a> </span><br />
<span style="font-size: x-small;">[facebook] <a href="https://www.facebook.com/deeplearningtalk/">https://www.facebook.com/deeplearningtalk/</a> </span><br />
<span style="font-size: x-small;"><br /></span><span style="font-size: x-small;">* T-Robotics의 글은 facebook과 rss reader로도 받아보실 수 있습니다. </span><br />
<span style="font-size: x-small;">[facebook] <a href="http://facebook.com/trobotics">http://facebook.com/trobotics</a> </span><br />
<br />
<div>
</div>
<br />
<div style="-webkit-text-stroke-width: 0px; color: black; font-family: "Malgun Gothic"; font-size: medium; font-style: normal; font-variant-caps: normal; font-variant-ligatures: normal; font-weight: 400; letter-spacing: normal; margin: 0px; orphans: 2; text-align: start; text-decoration-color: initial; text-decoration-style: initial; text-indent: 0px; text-transform: none; white-space: normal; widows: 2; word-spacing: 0px;">
<span style="font-size: x-small;">[rss] <a href="http://t-robotics.blogspot.kr/feeds/posts/default">http://t-robotics.blogspot.kr/feeds/posts/default</a></span></div>
Terryhttp://www.blogger.com/profile/10286616328664292643noreply@blogger.com0tag:blogger.com,1999:blog-2827101888996304029.post-34976972686241579162018-01-28T05:55:00.002-08:002018-02-16T19:57:56.918-08:00#31. Deep Learning: A Critical Appraisal이번에 리뷰한 논문은 논문이라기보단 칼럼에 가까운 글입니다. 딥러닝과 현재 딥러닝 연구들의 한계점들에 대해 이야기한 "Deep learning: A critical appraisal"이란 논문인데요, "왜 딥러닝을 이용해야하는가?", "딥러닝이 잘하는 것은 무엇이고 못하는 것은 무엇인가?", "인공지능은 얼마나 멀리있는가?" 등에 대해 이야기하려면 꼭 짚어봐야하는 내용들이죠.<br>
<br>
<a href="http://t-robotics.blogspot.com/2018/01/30-deep-learning-critical-appraisal.html#more">Read more »</a>Terryhttp://www.blogger.com/profile/10286616328664292643noreply@blogger.com0tag:blogger.com,1999:blog-2827101888996304029.post-19014622056893123562017-10-28T00:38:00.002-07:002018-02-16T19:57:46.796-08:00#30. ROC 커브 (+ AUC, Precision, Recall)Accuracy가 성능을 나타내는 전부는 아니란거 다들 알고 계시죠? 지난번엔 암환자 진단의 예를 통해 accuracy의 함정을 알아보고, precision과 recall에 대해서 설명을 했는데요,<br />
<br />
이번에는 이와 함께 많이 쓰이는 개념 중 하나인 ROC curve에 대해서 알아보았습니다. 특히 메디컬 페이퍼에선 신뢰성 있는 판단을 위해 ROC를 많이 이용하는데요, ROC란 무엇이고 도대체 왜 쓰는 것일까요? ROC에 대해 한번 알아보시죠.<br />
<br />
<div class="wp-caption aligncenter" id="attachment_25279" style="background-color: #f3f3f3; border-radius: 1px 3px 3px; border: 1px solid rgb(221, 221, 221); color: #222222; float: none; line-height: 21px; margin: 20px auto; max-width: 100%; outline: 0px; padding: 4px 0px 0px; text-align: center; vertical-align: baseline; width: 610px;">
<br />
<iframe allowfullscreen="" frameborder="0" height="252" src="https://www.youtube.com/embed/nMAtFhamoRY?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq" width="448"></iframe><br />
<b>[비디오]</b><br />
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEiyffAtD9DD1_fLyHO98yPn1x0Wn2NgxuC8wzPx7dluv6dB3VVrY0iLmDF0aid7Okzme2f_r_lreYtyQXc5yFztIIjm72LgH6SIU8cwTi-n7SFTz6qYB0rlL2YqEQpSx4wINzu5KfsvcU7I/s1600/Slide1.PNG" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" data-original-height="720" data-original-width="960" height="300" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEiyffAtD9DD1_fLyHO98yPn1x0Wn2NgxuC8wzPx7dluv6dB3VVrY0iLmDF0aid7Okzme2f_r_lreYtyQXc5yFztIIjm72LgH6SIU8cwTi-n7SFTz6qYB0rlL2YqEQpSx4wINzu5KfsvcU7I/s400/Slide1.PNG" width="400" /></a></div>
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEg5cVxJKigf860q3JTpsjlCdPALfonwn9esUbDixxtt7_m2A-ypfRKyIXQEYzxsXAP6ECxoD8cibg1Tq_0NoKeFw9VXhFwrrSsacyJs9g0azkwwXxosBhjkhfuDGlMkrz944pH6yVFKx5zU/s1600/Slide2.PNG" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" data-original-height="720" data-original-width="960" height="300" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEg5cVxJKigf860q3JTpsjlCdPALfonwn9esUbDixxtt7_m2A-ypfRKyIXQEYzxsXAP6ECxoD8cibg1Tq_0NoKeFw9VXhFwrrSsacyJs9g0azkwwXxosBhjkhfuDGlMkrz944pH6yVFKx5zU/s400/Slide2.PNG" width="400" /></a></div>
<div class="separator" style="clear: both; text-align: center;">
<br /></div>
<div class="separator" style="clear: both;">
<b>[슬라이드]</b></div>
<span style="font-size: x-small;"><br /></span></div>
<span style="font-size: x-small;"><br /></span><span style="font-size: x-small;">* 테리의 딥러닝 토크 </span><br />
<span style="font-size: x-small;">[youtube] <a href="https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq">https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq</a> </span><br />
<span style="font-size: x-small;">[facebook] <a href="https://www.facebook.com/deeplearningtalk/">https://www.facebook.com/deeplearningtalk/</a> </span><br />
<span style="font-size: x-small;"><br /></span><span style="font-size: x-small;">* T-Robotics의 글은 facebook과 rss reader로도 받아보실 수 있습니다. </span><br />
<span style="font-size: x-small;">[facebook] <a href="http://facebook.com/trobotics">http://facebook.com/trobotics</a> </span><br />
<span style="font-size: x-small;">[rss] <a href="http://t-robotics.blogspot.kr/feeds/posts/default">http://t-robotics.blogspot.kr/feeds/posts/default</a></span><br />
<div>
<br /></div>
Terryhttp://www.blogger.com/profile/10286616328664292643noreply@blogger.com0tag:blogger.com,1999:blog-2827101888996304029.post-24405058321917429372017-09-13T23:47:00.000-07:002017-09-13T23:47:02.671-07:00#29. Tensorflow vs. PyTorch딥러닝 공부를 시작하시는 분들의 큰 고민 중 하나가 짜장면이냐 짬뽕이냐가 아니라 Tensorflow냐 Pytorch냐일텐데요, 두가지를 가볍게 비교해 보았습니다. 결론은? 뭐든 그냥 손에 착 달라붙고 좀더 끌리는걸 하세요~ 저는 개인적으로 무엇을 선택했을까요...?^^<br />
<br />
[Reference] "PyTorch vs Tensorflow" https://medium.com/towards-data-science/pytorch-vs-tensorflow-spotting-the-difference-25c75777377b<br />
<br />
<div class="wp-caption aligncenter" id="attachment_25279" style="background-color: #f3f3f3; border-radius: 1px 3px 3px; border: 1px solid rgb(221, 221, 221); color: #222222; float: none; line-height: 21px; margin: 20px auto; max-width: 100%; outline: 0px; padding: 4px 0px 0px; text-align: center; vertical-align: baseline; width: 610px;">
<br />
<iframe allowfullscreen="" frameborder="0" height="252" src="https://www.youtube.com/embed/tNWatDufzDk?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq" width="448"></iframe><br />
<b>[비디오]</b><br />
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEgWixJLQEuNdwY-MYvsZvhS8ySZHplmcxlapkUs2OVYk96qDZWVgN0us09AjMROmRFDF3cfRf6NTCRFaJ6OzJ0rgibaObSYlBnujaUEnj-coMcleO_Y3nKSLN4GSsoNMr5Ltom1Jv7BESHK/s1600/Slide1.PNG" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" data-original-height="720" data-original-width="960" height="300" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEgWixJLQEuNdwY-MYvsZvhS8ySZHplmcxlapkUs2OVYk96qDZWVgN0us09AjMROmRFDF3cfRf6NTCRFaJ6OzJ0rgibaObSYlBnujaUEnj-coMcleO_Y3nKSLN4GSsoNMr5Ltom1Jv7BESHK/s400/Slide1.PNG" width="400" /></a></div>
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEhM1NPwIPKR-RcAF-9_OwcYgLmjoujZAgP5qlhY7_RYyihD3gBBPKykZA_mphoZZ59jijZmf_hSMmqWNMp7Jgu4gw0Yp0QMgn_fJTrHE3pgY_rhQMyqxRWWMD4HgSNbDEU9c60asBcGBOYR/s1600/Slide2.PNG" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" data-original-height="720" data-original-width="960" height="300" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEhM1NPwIPKR-RcAF-9_OwcYgLmjoujZAgP5qlhY7_RYyihD3gBBPKykZA_mphoZZ59jijZmf_hSMmqWNMp7Jgu4gw0Yp0QMgn_fJTrHE3pgY_rhQMyqxRWWMD4HgSNbDEU9c60asBcGBOYR/s400/Slide2.PNG" width="400" /></a></div>
<div class="separator" style="clear: both;">
<b>[슬라이드]</b></div>
<span style="font-size: x-small;"><br /></span></div>
<span style="font-size: x-small;"><br /></span><span style="font-size: x-small;">* 테리의 딥러닝 토크 </span><br />
<span style="font-size: x-small;">[youtube] <a href="https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq">https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq</a> </span><br />
<span style="font-size: x-small;">[facebook] <a href="https://www.facebook.com/deeplearningtalk/">https://www.facebook.com/deeplearningtalk/</a> </span><br />
<span style="font-size: x-small;"><br /></span><span style="font-size: x-small;">* T-Robotics의 글은 facebook과 rss reader로도 받아보실 수 있습니다. </span><br />
<span style="font-size: x-small;">[facebook] <a href="http://facebook.com/trobotics">http://facebook.com/trobotics</a> </span><br />
<span style="font-size: x-small;">[rss] <a href="http://t-robotics.blogspot.kr/feeds/posts/default">http://t-robotics.blogspot.kr/feeds/posts/default</a></span><br />
<div>
<br /></div>
Terryhttp://www.blogger.com/profile/10286616328664292643noreply@blogger.com0tag:blogger.com,1999:blog-2827101888996304029.post-70861915346617822052017-09-10T07:24:00.003-07:002017-09-10T07:24:45.877-07:00#28. Understanding Black-box Predictions via Influence Functions Tensorflow KR에서 진행하고 있는 논문읽기 모임 PR12에서 발표한 저의 네번째 발표입니다. 이번에는 ICML2017에서 베스트페이퍼상을 받은 "딥러닝의 결과를 어떻게 이해할 수 있는가"에 대한 논문을 리뷰해보았습니다.<br />
<br />
딥러닝은 성능은 좋지만 왜 그게 잘되는지 모르는, 그야말로 "블랙박스"와 같은 모델인데요, 이 논문에서는 '만약 A라는 트레이닝 데이터가 없다면 어떤 변화가 일어날까?', 'B라는 테스트 이미지에 가장 결정적인 영향을 주는 트레이닝 데이터는 무엇일까?'와 같은 질문에 대해 influence function이라는 것을 도입해서 해결하려 하였습니다.<br />
<br />
"Understanding Black-box Predictions via Influence Functions", Pang Wei Koh and Percy Liang, 2017<br />
[Link] <a href="http://proceedings.mlr.press/v70/koh17a.html">http://proceedings.mlr.press/v70/koh17a.html</a><br />
<br />
[YouTube] <a href="https://www.youtube.com/watch?v=xlmlY8WHjkU&index=30&list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq">https://www.youtube.com/watch?v=xlmlY8WHjkU&index=30&list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq</a><br />
[Slides] <a href="https://www.slideshare.net/secret/1Bqa2zTPde1Uc4">https://www.slideshare.net/secret/1Bqa2zTPde1Uc4</a><br />
<div class="wp-caption aligncenter" id="attachment_25279" style="background-color: #f3f3f3; border-radius: 1px 3px 3px; border: 1px solid rgb(221, 221, 221); float: none; line-height: 21px; margin: 20px auto; max-width: 100%; outline: 0px; padding: 4px 0px 0px; text-align: center; vertical-align: baseline; width: 610px;">
<div style="text-align: start;">
<br /></div>
<div style="text-align: start;">
<br /></div>
<div style="color: #222222;">
<iframe allowfullscreen="" frameborder="0" height="252" src="https://www.youtube.com/embed/xlmlY8WHjkU?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq" width="448"></iframe><br /></div>
<div style="color: #222222;">
<b>[비디오]</b></div>
<div style="color: #222222;">
<br /></div>
<div class="separator" style="clear: both; color: #222222;">
<br /></div>
<div class="separator" style="clear: both; color: #222222;">
<iframe allowfullscreen="" frameborder="0" height="380" marginheight="0" marginwidth="0" scrolling="no" src="https://www.slideshare.net/slideshow/embed_code/key/1Bqa2zTPde1Uc4" style="border-color: rgb(204, 204, 204); border-style: solid; border-width: 1px; margin-bottom: 5px; max-width: 100%;" width="448"></iframe></div>
<div style="color: #222222;">
<b>[슬라이드]</b></div>
<div style="color: #222222;">
<br /></div>
</div>
<span style="font-size: x-small;"><span style="background-color: #fefdfa; color: #333333; font-family: "arial" , "helvetica" , sans-serif; line-height: 14px; text-align: justify;"><br /></span></span><span style="font-size: x-small;"><span style="background-color: #fefdfa; color: #333333; font-family: "arial" , "helvetica" , sans-serif; line-height: 14px; text-align: justify;">* 테리의 딥러닝 토크</span></span><br />
<span style="font-size: x-small;"><span style="background-color: #fefdfa; color: #333333; font-family: "arial" , "helvetica" , sans-serif; line-height: 14px; text-align: justify;">[youtube] <a href="https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq">https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq</a> </span></span><br />
<span style="font-size: x-small;"><span style="background-color: #fefdfa; color: #333333; font-family: "arial" , "helvetica" , sans-serif; line-height: 14px; text-align: justify;">[facebook] <a href="https://www.facebook.com/deeplearningtalk/">https://www.facebook.com/deeplearningtalk/</a></span></span><br />
<span style="font-size: x-small;"><span style="background-color: #fefdfa; color: #333333; font-family: "arial" , "helvetica" , sans-serif; line-height: 14px; text-align: justify;"><br /></span><span style="background-color: #fefdfa; color: #333333; font-family: "arial" , "helvetica" , sans-serif; line-height: 14px; text-align: justify;">* T-Robotics의 글은 facebook과 rss reader로도 받아보실 수 있습니다.</span></span><br />
<div class="separator" style="background-color: #fefdfa; clear: both; color: #333333; font-family: verdana, geneva, sans-serif; line-height: 18px; text-align: justify;">
<span style="font-family: "arial" , "helvetica" , sans-serif; font-size: x-small;">[<span class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="189cb4ff-7cae-4873-8436-d55ddbe2c9f6" id="b095b228-3a62-4807-8be7-f313dc7b6aa8"><gs class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="61cefc52-e6f8-427a-999c-d8e2bd5d8157" id="9028c2f4-078e-43ba-9375-3f608197542a"><gs class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="1db50239-7e8d-4f62-a464-d9e4af1d255e" id="6c27a77b-a00b-4dea-ad4e-58c92c705bf7">facebook</gs></gs></span><span class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="6c2e595c-3435-49db-a163-72e32681ebb2" id="a9fd5842-c391-483e-bb95-bf8f2fa72c85">]</span> <a href="http://facebook.com/trobotics" style="color: #7d181e;">http://facebook.com/trobotics</a></span></div>
<div style="font-family: "맑은 고딕"; margin: 0px;">
<span style="font-size: x-small;"><span style="background-color: #fefdfa; color: #333333; font-family: "verdana" , "geneva" , sans-serif; text-align: justify;">[</span><span class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="9fbdb27b-fee2-46c4-a36e-dc4f19ead275" id="1ff5bd24-5148-4c66-b842-46763eae0686" style="background-color: #fefdfa; color: #333333; font-family: "verdana" , "geneva" , sans-serif; text-align: justify;"><gs class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="df400c24-2e15-4c1e-ad29-f5e8fc8f4881" id="d7151f55-e778-4855-ab9f-bd47817d510e"><gs class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="caf4142c-4d49-46d3-a002-1f57a5fa5a22" id="13516183-5b6c-4dbf-9040-b5088aa14c77">rss</gs></gs></span><span style="background-color: #fefdfa; color: #333333; font-family: "verdana" , "geneva" , sans-serif; text-align: justify;">] </span><a href="http://t-robotics.blogspot.kr/feeds/posts/default" style="background-color: #fefdfa; color: #7d181e; font-family: arial, helvetica, sans-serif; line-height: 14px; text-align: justify;">http://t-robotics.blogspot.kr/feeds/posts/default</a></span><br />
<div>
<br /></div>
<div>
</div>
</div>
Terryhttp://www.blogger.com/profile/10286616328664292643noreply@blogger.com0tag:blogger.com,1999:blog-2827101888996304029.post-89101490851348576632017-08-13T02:56:00.001-07:002017-08-13T02:56:48.937-07:00#27. Cross entropy와 KL divergence<br />
혹시 "classification을 위해 cross-entropy를 loss function으로 사용합니다" 라든지, "분포의 차이를 줄이기 위해 KL-divergence를 최소화시킵니다"와 같은 이야기를 들어보신 적 있으신가요?<br />
<br />
정보의 양을 따질 때, 혹은 확률 분포들의 차이를 따질 때 중요하게 사용되는 개념 중 하나가 바로 Entropy인데요, 오늘은 엔트로피란 무엇인지에 대해 쉬운 예제와 함께 살펴봤습니다.<br />
<br />
결론을 말씀드리자면 "엔트로피는 정보를 최적으로 인코딩하기 위해 필요한 bit수이며, 각 label들의 확률 분포의 함수"입니다.<br />
<br />
<div class="wp-caption aligncenter" id="attachment_25279" style="background-color: #f3f3f3; border-radius: 1px 3px 3px; border: 1px solid rgb(221, 221, 221); color: #222222; float: none; line-height: 21px; margin: 20px auto; max-width: 100%; outline: 0px; padding: 4px 0px 0px; text-align: center; vertical-align: baseline; width: 610px;">
<br />
<iframe allowfullscreen="" frameborder="0" height="252" src="https://www.youtube.com/embed/uMYhthKw1PU?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq" width="448"></iframe><br />
<b>[비디오]</b><br />
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEgXFZ58rKiy-AnEqmAq633Gdgn6wWKa01EWPrMfJmjzdb75pzAaOb3f9seK3nm9JtJ9g5wXK_-Noe6qYYsRdfbKkAGKKQg5RB5gYer8wDaWwZyvQHay9mOzomufkmv2wcWPAad3voxR8FDp/s1600/Slide1.PNG" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" data-original-height="720" data-original-width="960" height="300" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEgXFZ58rKiy-AnEqmAq633Gdgn6wWKa01EWPrMfJmjzdb75pzAaOb3f9seK3nm9JtJ9g5wXK_-Noe6qYYsRdfbKkAGKKQg5RB5gYer8wDaWwZyvQHay9mOzomufkmv2wcWPAad3voxR8FDp/s400/Slide1.PNG" width="400" /></a></div>
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEhvVBJzY3kPQC7u2zI7Hp4iyNOz9Q8iG9Z9YNKGZV2kvctqYtf9yCEAmA2Y4gnzqKDYMV3JBvu11BoPQ81cIMkAM6CT8za7oaBWB-qq3aqnC_zJqRLVPAzbkf8ycrjwXaCQpwAu9XthQ6jH/s1600/Slide2.PNG" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" data-original-height="720" data-original-width="960" height="300" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEhvVBJzY3kPQC7u2zI7Hp4iyNOz9Q8iG9Z9YNKGZV2kvctqYtf9yCEAmA2Y4gnzqKDYMV3JBvu11BoPQ81cIMkAM6CT8za7oaBWB-qq3aqnC_zJqRLVPAzbkf8ycrjwXaCQpwAu9XthQ6jH/s400/Slide2.PNG" width="400" /></a></div>
<div class="separator" style="clear: both; text-align: center;">
<b>[슬라이드]</b></div>
<span style="font-size: x-small;"><br /></span></div>
<span style="font-size: x-small;"><br /></span> <span style="font-size: x-small;">* 테리의 딥러닝 토크 </span><br />
<span style="font-size: x-small;">[youtube] <a href="https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq">https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq</a> [facebook] <a href="https://www.facebook.com/deeplearningtalk/">https://www.facebook.com/deeplearningtalk/</a> </span><br />
<span style="font-size: x-small;"><br /></span> <span style="font-size: x-small;">* T-Robotics의 글은 facebook과 rss reader로도 받아보실 수 있습니다. </span><br />
<span style="font-size: x-small;">[facebook] <a href="http://facebook.com/trobotics">http://facebook.com/trobotics</a> </span><br />
<span style="font-size: x-small;">[rss] <a href="http://t-robotics.blogspot.kr/feeds/posts/default">http://t-robotics.blogspot.kr/feeds/posts/default</a></span><br />
<br />Terryhttp://www.blogger.com/profile/10286616328664292643noreply@blogger.com0tag:blogger.com,1999:blog-2827101888996304029.post-42487306427590193992017-08-05T22:13:00.001-07:002017-08-05T22:13:30.790-07:00#26. 정보량을 나타내는 엔트로피 (Entropy)<br />
혹시 "classification을 위해 cross-entropy를 loss function으로 사용합니다" 라든지, "분포의 차이를 줄이기 위해 KL-divergence를 최소화시킵니다"와 같은 이야기를 들어보신 적 있으신가요?<br />
<br />
정보의 양을 따질 때, 혹은 확률 분포들의 차이를 따질 때 중요하게 사용되는 개념 중 하나가 바로 Entropy인데요, 오늘은 엔트로피란 무엇인지에 대해 쉬운 예제와 함께 살펴봤습니다.<br />
<br />
결론을 말씀드리자면 "엔트로피는 정보를 최적으로 인코딩하기 위해 필요한 bit수이며, 각 label들의 확률 분포의 함수"입니다.<br />
<br />
<div class="wp-caption aligncenter" id="attachment_25279" style="background-color: #f3f3f3; border-radius: 1px 3px 3px; border: 1px solid rgb(221, 221, 221); color: #222222; float: none; line-height: 21px; margin: 20px auto; max-width: 100%; outline: 0px; padding: 4px 0px 0px; text-align: center; vertical-align: baseline; width: 610px;">
<br />
<iframe allowfullscreen="" frameborder="0" height="252" src="https://www.youtube.com/embed/zJmbkp9TCXY?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq" width="448"></iframe><br />
<b>[비디오]</b><br />
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEiolz6OctEUm5LcdQDKxH77lNNcBswG0FhFon9XDT-zORF9WkUurg0RuRnsAc7GdmNUHb7XEu2KXWV0PHUazj7bO2WQwxqUZsWVyRoJ7ysq3yo_yzjSC48uWdTJ2a5LBzh8sNTwA9jVnyPx/s1600/Slide1.PNG" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" data-original-height="720" data-original-width="960" height="300" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEiolz6OctEUm5LcdQDKxH77lNNcBswG0FhFon9XDT-zORF9WkUurg0RuRnsAc7GdmNUHb7XEu2KXWV0PHUazj7bO2WQwxqUZsWVyRoJ7ysq3yo_yzjSC48uWdTJ2a5LBzh8sNTwA9jVnyPx/s400/Slide1.PNG" width="400" /></a></div>
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEh5VLhdRjHnkrVzk24wYXDYu9GmfMSSO5VFJRNPTo1j1Ek2K3n-J8GLN2Xq3UfLj386hZv-sGq1ikRtXPbq6bKXZSaurdAEetuJeQlUP3GSap8kEaC55gzIiREaz-qKWQTwvWG9t0hsxkOB/s1600/Slide2.PNG" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" data-original-height="720" data-original-width="960" height="300" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEh5VLhdRjHnkrVzk24wYXDYu9GmfMSSO5VFJRNPTo1j1Ek2K3n-J8GLN2Xq3UfLj386hZv-sGq1ikRtXPbq6bKXZSaurdAEetuJeQlUP3GSap8kEaC55gzIiREaz-qKWQTwvWG9t0hsxkOB/s400/Slide2.PNG" width="400" /></a></div>
<div class="separator" style="clear: both; text-align: center;">
<b>[슬라이드]</b></div>
<span style="font-size: x-small;"><br /></span></div>
<span style="font-size: x-small;"><br /></span> <span style="font-size: x-small;">* 테리의 딥러닝 토크 </span><br />
<span style="font-size: x-small;">[youtube] <a href="https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq">https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq</a> [facebook] <a href="https://www.facebook.com/deeplearningtalk/">https://www.facebook.com/deeplearningtalk/</a> </span><br />
<span style="font-size: x-small;"><br /></span> <span style="font-size: x-small;">* T-Robotics의 글은 facebook과 rss reader로도 받아보실 수 있습니다. </span><br />
<span style="font-size: x-small;">[facebook] <a href="http://facebook.com/trobotics">http://facebook.com/trobotics</a> </span><br />
<span style="font-size: x-small;">[rss] <a href="http://t-robotics.blogspot.kr/feeds/posts/default">http://t-robotics.blogspot.kr/feeds/posts/default</a></span><br />
<br />Terryhttp://www.blogger.com/profile/10286616328664292643noreply@blogger.com0tag:blogger.com,1999:blog-2827101888996304029.post-26396046189019100232017-07-23T07:39:00.001-07:002017-07-23T07:43:14.120-07:00#25. Learning with side information through modality hallucination (2016) Tensorflow KR에서 진행하고 있는 논문읽기 모임 PR12에서 발표한 저의 세번째 발표입니다. 이번에는 "다양한 재료들을 어떻게 학습에 사용할 수 있을까"에 대한 내용을 발표하였습니다.<br />
<br />
전형적인 supervised learning setting에선 이미지와 레이블을 주고 그것의 매핑을 학습하는데요, 사실 이미지 말고도 우리가 줄 수 있는 정보가 더 많을 수도 있습니다. 예를 들면 Kinect를 이용한 depth image 처럼 말이죠.<br />
<br />
하지만 문제는 depth image를 가지고 트레이닝을 할 경우 테스트 때 (런타임 때)도 depth image가 있어야지 prediction이 가능하다는 단점이 있습니다. 있는 재료를 활용하려다 더 불편함만 얻게되는 결과지요. 근데 그렇다고 트레이닝 때 더 있는 재료를 활용 안하는 것도 최선은 아닌 것 같고요...<br />
<br />
이 논문은 이처럼 트레이닝 때 이용한 modality(인풋 소스)가 테스트 때는 모두 이용할 수 없을 때 이것을 할루시네이션처럼 "복사"함으로써 해결하는 방법을 제시했습니다. 수학도 없고 들어보시면 재밌으실 겁니다 ㅎㅎ 그럼 즐겁게 들어주세요~<br />
<br />
"Learning with side information through modality hallucination", J. Hoffman et al., 2017.<br />
[Link] <a href="http://www.cv-foundation.org/openaccess/content_cvpr_2016/html/Hoffman_Learning_With_Side_CVPR_2016_paper.html">http://www.cv-foundation.org/openaccess/content_cvpr_2016/html/Hoffman_Learning_With_Side_CVPR_2016_paper.html</a><br />
<div style="text-align: center;">
<div style="text-align: left;">
</div>
</div>
<div style="text-align: center;">
<div style="text-align: left;">
<br /></div>
<div style="text-align: left;">
[YouTube] <a href="https://www.youtube.com/watch?v=KdRo7ATNs9g&t=629s&index=27&list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq&spfreload=10">https://www.youtube.com/watch?v=KdRo7ATNs9g&t=629s&index=27&list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq&spfreload=10</a></div>
</div>
<div style="text-align: center;">
<div style="text-align: left;">
[Slides] <a href="https://www.slideshare.net/TerryTaewoongUm/learning-with-side-information-through-modality-hallucination-2016">https://www.slideshare.net/TerryTaewoongUm/learning-with-side-information-through-modality-hallucination-2016</a></div>
</div>
<div class="wp-caption aligncenter" id="attachment_25279" style="background-color: #f3f3f3; border-radius: 1px 3px 3px; border: 1px solid rgb(221, 221, 221); float: none; line-height: 21px; margin: 20px auto; max-width: 100%; outline: 0px; padding: 4px 0px 0px; text-align: center; vertical-align: baseline; width: 610px;">
<div style="text-align: start;">
<br /></div>
<div style="text-align: start;">
<br /></div>
<div style="color: #222222;">
<iframe allowfullscreen="" frameborder="0" height="252" src="https://www.youtube.com/embed/Y1loN3Sc4Dk?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq" width="448"></iframe><br /></div>
<div style="color: #222222;">
<b>[비디오]</b></div>
<div style="color: #222222;">
<br /></div>
<div class="separator" style="clear: both; color: #222222;">
<br /></div>
<div class="separator" style="clear: both; color: #222222;">
<iframe allowfullscreen="" frameborder="0" height="320" marginheight="0" marginwidth="0" scrolling="no" src="//www.slideshare.net/slideshow/embed_code/key/HHguc5R54ttxcA" style="border-width: 1px; border: 1px solid #ccc; margin-bottom: 5px; max-width: 100%;" width="448"> </iframe> </div>
<div style="margin-bottom: 5px;">
<strong> <a href="https://www.slideshare.net/TerryTaewoongUm/learning-with-side-information-through-modality-hallucination-2016" target="_blank" title="Learning with side information through modality hallucination (2016)">Learning with side information through modality hallucination (2016)</a> </strong> from <strong><a href="https://www.slideshare.net/TerryTaewoongUm" target="_blank">Terry Taewoong Um</a></strong> </div>
<div style="color: #222222;">
<b>[슬라이드]</b></div>
</div>
<span style="font-size: x-small;"><br /></span> <span style="font-size: x-small;">* T-Robotics의 글은 facebook과 rss reader로도 받아보실 수 있습니다.</span><br />
<span style="font-size: x-small;">[facebook] <a href="http://facebook.com/trobotics">http://facebook.com/trobotics</a></span><br />
<span style="font-size: x-small;">[rss] <a href="http://t-robotics.blogspot.kr/feeds/posts/default">http://t-robotics.blogspot.kr/feeds/posts/default</a></span><br />
<br />
<br />Terryhttp://www.blogger.com/profile/10286616328664292643noreply@blogger.com0tag:blogger.com,1999:blog-2827101888996304029.post-18107158618259618902017-07-08T00:20:00.000-07:002017-07-08T00:20:34.036-07:00#24. Intra- / Inter-Class Variability여러분들이 만약 classification(분류)을 머신러닝을 통해 하고싶다면, 그냥 데이터만 다 때려넣고 머신러닝을 돌리면 될까요? 아마도 데이터의 질에 따라 될 수도 있고, 안될 수도 있을텐데요, 그렇다면 데이터의 '질'이란 과연 무엇일까요?<br />
<br />
Classification에서 나의 데이터 혹은 나의 피쳐의 상태를 가장 잘 표현해주는 단어는 아마도 '클래스 내부의 분산이 어떠냐(intra-class variability)', 그리고 '클래스 간 분신이 어떠냐(inter-class variability)'일 것입니다. 이게 뭐냐고요? 영상을 보시죠 ^^<br />
<br />
<div class="wp-caption aligncenter" id="attachment_25279" style="background-color: #f3f3f3; border-radius: 1px 3px 3px; border: 1px solid rgb(221, 221, 221); color: #222222; float: none; line-height: 21px; margin: 20px auto; max-width: 100%; outline: 0px; padding: 4px 0px 0px; text-align: center; vertical-align: baseline; width: 610px;">
<br />
<iframe allowfullscreen="" frameborder="0" height="252" src="https://www.youtube.com/embed/kiPo5ACr9xw?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq" width="448"></iframe><br />
<b>[비디오]</b><br />
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEiq3SFUq_G0NasOwShaB40MY77JIuw4GQWonAW9vAa-mwxqYhq8xf1w8KQrxVmEi-Of878iM9fR0p4H3X4g-yUin00EfrhTVCQhWFTKKhzxMElAErgoWJumFLSLE6wNmSyDqAuJUnJnokoK/s1600/Slide1.PNG" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" data-original-height="720" data-original-width="960" height="300" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEiq3SFUq_G0NasOwShaB40MY77JIuw4GQWonAW9vAa-mwxqYhq8xf1w8KQrxVmEi-Of878iM9fR0p4H3X4g-yUin00EfrhTVCQhWFTKKhzxMElAErgoWJumFLSLE6wNmSyDqAuJUnJnokoK/s400/Slide1.PNG" width="400" /></a></div>
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEjwnxO18xQKZDC-AbNplp7UxREHTAjtScPi35z4M1Q6YW5aHAPKYr42P0jA7OpIQN6GWL4ZPMVqfEXeG1SQMyD7jZVTxnzFk58vnrNo5h3U04LRl7tWShPKd_PO3tgWAUhYk61JH-h4jVkJ/s1600/Slide2.PNG" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" data-original-height="720" data-original-width="960" height="300" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEjwnxO18xQKZDC-AbNplp7UxREHTAjtScPi35z4M1Q6YW5aHAPKYr42P0jA7OpIQN6GWL4ZPMVqfEXeG1SQMyD7jZVTxnzFk58vnrNo5h3U04LRl7tWShPKd_PO3tgWAUhYk61JH-h4jVkJ/s400/Slide2.PNG" width="400" /></a></div>
<div class="separator" style="clear: both;">
<br /></div>
<div class="separator" style="clear: both;">
<b>[슬라이드]</b></div>
<span style="font-size: x-small;"><br /></span></div>
<span style="font-size: x-small;"><br /></span>
<span style="font-size: x-small;">* 테리의 딥러닝 토크 </span><br />
<span style="font-size: x-small;">[youtube] <a href="https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq">https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq</a> [facebook] <a href="https://www.facebook.com/deeplearningtalk/">https://www.facebook.com/deeplearningtalk/</a> </span><br />
<span style="font-size: x-small;"><br /></span>
<span style="font-size: x-small;">* T-Robotics의 글은 facebook과 rss reader로도 받아보실 수 있습니다. </span><br />
<span style="font-size: x-small;">[facebook] <a href="http://facebook.com/trobotics">http://facebook.com/trobotics</a> </span><br />
<span style="font-size: x-small;">[rss] <a href="http://t-robotics.blogspot.kr/feeds/posts/default">http://t-robotics.blogspot.kr/feeds/posts/default</a></span><br />
<br />Terryhttp://www.blogger.com/profile/10286616328664292643noreply@blogger.com0tag:blogger.com,1999:blog-2827101888996304029.post-84282990499627219092017-06-29T17:16:00.000-07:002017-06-29T17:27:41.236-07:00#23. RNN & LSTM딥러닝의 양대 축이라고 하면 Convolutional Neural Network (CNN)와 Recurrent Neural Network (RNN)를 꼽을 수 있을텐데요, 많은 분들이 RNN을 쓸 때 주로 이용되는 Long Short-term Memory (LSTM)의 이해에 어려움을 겪으시는 것 같습니다.<br />
<br />
그래서 준비했습니다. RNN과 LSTM 이해를 위한 완벽가이드! 금쪽같은 블로그인 Colah의 블로그 내용을 기반으로 RNN과 LSTM의 내부 메커니즘과 수식을 설명해보았습니다. 그럼 즐겁게 들어주세요~<br />
<br />
[Link] Colah의 블로그, <a href="http://colah.github.io/posts/2015-08-Understanding-LSTMs/">http://colah.github.io/posts/2015-08-Understanding-LSTMs/</a><br />
<br />
<div class="wp-caption aligncenter" id="attachment_25279" style="background-color: #f3f3f3; border-radius: 1px 3px 3px; border: 1px solid rgb(221, 221, 221); color: #222222; float: none; line-height: 21px; margin: 20px auto; max-width: 100%; outline: 0px; padding: 4px 0px 0px; text-align: center; vertical-align: baseline; width: 610px;">
<br />
<iframe allowfullscreen="" frameborder="0" height="252" src="https://www.youtube.com/embed/SoNtAjxA3Jo?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq" width="448"></iframe><br />
<b>[비디오]</b><br />
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEiE2W0nzvZJo2k7ln5L4UJZw3El-3WyzwKBojEzRLc9PSZ7GCRf0ShAY5raIfiXnAWf3-D2bnC2vsHqsP8y4f8IZR2ijfw5G_L15pkdT5sXdlGU8HzWh7bt4MZ3dkfNcd1xG1Jh4Y4bJS7M/s1600/Slide1.PNG" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" data-original-height="720" data-original-width="960" height="300" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEiE2W0nzvZJo2k7ln5L4UJZw3El-3WyzwKBojEzRLc9PSZ7GCRf0ShAY5raIfiXnAWf3-D2bnC2vsHqsP8y4f8IZR2ijfw5G_L15pkdT5sXdlGU8HzWh7bt4MZ3dkfNcd1xG1Jh4Y4bJS7M/s400/Slide1.PNG" width="400" /></a></div>
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEh5GdbQCm88FR7jpZRLLQqatSVgz6XNE6uynl_ryqMkgHyPPlDhVKhBFCHA9rWWwydIPFnoyPd0ZbcxXrq6qhMkkqEXUiKOg90zh5leUvIug8349NrbOB64DRuKvm1lz3k3iPrKHeY2jcTC/s1600/Slide2.PNG" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" data-original-height="720" data-original-width="960" height="300" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEh5GdbQCm88FR7jpZRLLQqatSVgz6XNE6uynl_ryqMkgHyPPlDhVKhBFCHA9rWWwydIPFnoyPd0ZbcxXrq6qhMkkqEXUiKOg90zh5leUvIug8349NrbOB64DRuKvm1lz3k3iPrKHeY2jcTC/s400/Slide2.PNG" width="400" /></a></div>
<div class="separator" style="clear: both;">
<b>[슬라이드]</b></div>
<br /></div>
<span style="font-size: x-small;"><span style="background-color: #fefdfa; color: #333333; font-family: "arial" , "helvetica" , sans-serif; line-height: 14px; text-align: justify;"><br /></span></span><span style="font-size: x-small;"><span style="background-color: #fefdfa; color: #333333; font-family: "arial" , "helvetica" , sans-serif; line-height: 14px; text-align: justify;">* 테리의 딥러닝 토크</span></span><br />
<span style="font-size: x-small;"><span style="background-color: #fefdfa; color: #333333; font-family: "arial" , "helvetica" , sans-serif; line-height: 14px; text-align: justify;">[youtube] <a href="https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq">https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq</a> </span></span><br />
<span style="font-size: x-small;"><span style="background-color: #fefdfa; color: #333333; font-family: "arial" , "helvetica" , sans-serif; line-height: 14px; text-align: justify;">[facebook] <a href="https://www.facebook.com/deeplearningtalk/">https://www.facebook.com/deeplearningtalk/</a></span></span><br />
<span style="font-size: x-small;"><span style="background-color: #fefdfa; color: #333333; font-family: "arial" , "helvetica" , sans-serif; line-height: 14px; text-align: justify;"><br /></span><span style="background-color: #fefdfa; color: #333333; font-family: "arial" , "helvetica" , sans-serif; line-height: 14px; text-align: justify;">* T-Robotics의 글은 facebook과 rss reader로도 받아보실 수 있습니다.</span></span><br />
<div class="separator" style="background-color: #fefdfa; clear: both; color: #333333; font-family: verdana, geneva, sans-serif; line-height: 18px; text-align: justify;">
<span style="font-family: "arial" , "helvetica" , sans-serif; font-size: x-small;">[<span class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="189cb4ff-7cae-4873-8436-d55ddbe2c9f6" id="b095b228-3a62-4807-8be7-f313dc7b6aa8"><gs class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="61cefc52-e6f8-427a-999c-d8e2bd5d8157" id="9028c2f4-078e-43ba-9375-3f608197542a"><gs class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="1db50239-7e8d-4f62-a464-d9e4af1d255e" id="6c27a77b-a00b-4dea-ad4e-58c92c705bf7">facebook</gs></gs></span><span class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="6c2e595c-3435-49db-a163-72e32681ebb2" id="a9fd5842-c391-483e-bb95-bf8f2fa72c85">]</span> <a href="http://facebook.com/trobotics" style="color: #7d181e; text-decoration: none;">http://facebook.com/trobotics</a></span></div>
<div style="-webkit-text-stroke-width: 0px; color: black; font-family: "맑은 고딕"; font-size: medium; font-style: normal; font-variant-caps: normal; font-variant-ligatures: normal; font-weight: normal; letter-spacing: normal; margin: 0px; orphans: 2; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; widows: 2; word-spacing: 0px;">
<span style="font-size: x-small;"><span style="background-color: #fefdfa; color: #333333; font-family: "verdana" , "geneva" , sans-serif; text-align: justify;">[</span><span class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="9fbdb27b-fee2-46c4-a36e-dc4f19ead275" id="1ff5bd24-5148-4c66-b842-46763eae0686" style="background-color: #fefdfa; color: #333333; font-family: "verdana" , "geneva" , sans-serif; text-align: justify;"><gs class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="df400c24-2e15-4c1e-ad29-f5e8fc8f4881" id="d7151f55-e778-4855-ab9f-bd47817d510e"><gs class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="caf4142c-4d49-46d3-a002-1f57a5fa5a22" id="13516183-5b6c-4dbf-9040-b5088aa14c77">rss</gs></gs></span><span style="background-color: #fefdfa; color: #333333; font-family: "verdana" , "geneva" , sans-serif; text-align: justify;">] </span><a href="http://t-robotics.blogspot.kr/feeds/posts/default" style="background-color: #fefdfa; color: #7d181e; font-family: arial, helvetica, sans-serif; line-height: 14px; text-align: justify; text-decoration: none;">http://t-robotics.blogspot.kr/feeds/posts/default</a></span></div>
Terryhttp://www.blogger.com/profile/10286616328664292643noreply@blogger.com0tag:blogger.com,1999:blog-2827101888996304029.post-23617090278741959122017-06-04T14:46:00.001-07:002017-06-29T17:03:29.465-07:00#22. On Human Motion Prediction using RNNs (2017) Tensorflow KR에서 진행하고 있는 논문읽기 모임 PR12에서의 저의 두번째 발표입니다. 이번에는 자연어에 대한 Seq2Seq (sequence-to-sequence) learning의 확장으로서 모션을 예측하는 motion forecasting에 대해 이야기해 보았는데요, 예를 들면 사람의 배트스윙 모션을 보고 그 다음의 모션들을 예측할 수 있는지에 대한 문제입니다. 아직은 1초 내외의 짧은 순간만을 예측하고 있네요. 그럼 즐겁게 들어주세요~!<br />
<div style="text-align: center;">
<div style="text-align: left;">
<br /></div>
</div>
<div style="text-align: center;">
<div style="text-align: left;">
"On Human Motion Prediction using Recurrent Neural Networks", J. Martinez, M. J. Black and J. Romero, 2017.</div>
</div>
<div style="text-align: center;">
<div style="text-align: left;">
[Link] <a href="https://arxiv.org/abs/1705.02445">https://arxiv.org/abs/1705.02445</a></div>
</div>
<div style="text-align: center;">
<div style="text-align: left;">
<br /></div>
<div style="text-align: left;">
[YouTube] <a href="https://www.youtube.com/watch?v=Y1loN3Sc4Dk&index=23&list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq">https://www.youtube.com/watch?v=Y1loN3Sc4Dk&index=23&list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq</a></div>
</div>
<div style="text-align: center;">
<div style="text-align: left;">
[Slides] <a href="https://www.slideshare.net/TerryTaewoongUm/human-motion-forecasting-generation-with-rnns">https://www.slideshare.net/TerryTaewoongUm/human-motion-forecasting-generation-with-rnns</a></div>
</div>
<div class="wp-caption aligncenter" id="attachment_25279" style="background-color: #f3f3f3; border-radius: 1px 3px 3px; border: 1px solid rgb(221, 221, 221); float: none; line-height: 21px; margin: 20px auto; max-width: 100%; outline: 0px; padding: 4px 0px 0px; text-align: center; vertical-align: baseline; width: 610px;">
<div style="text-align: start;">
<br /></div>
<div style="text-align: start;">
<br /></div>
<div style="color: #222222;">
<iframe allowfullscreen="" frameborder="0" height="252" src="https://www.youtube.com/embed/Y1loN3Sc4Dk?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq" width="448"></iframe><br /></div>
<div style="color: #222222;">
<b>[비디오]</b></div>
<div style="color: #222222;">
<br /></div>
<div class="separator" style="clear: both; color: #222222;">
<br /></div>
<div class="separator" style="clear: both; color: #222222;">
<iframe allowfullscreen="" frameborder="0" height="380" marginheight="0" marginwidth="0" scrolling="no" src="https://www.slideshare.net/slideshow/embed_code/key/K1Hme1U1gJVRZ8" style="border-color: rgb(204, 204, 204); border-style: solid; border-width: 1px; margin-bottom: 5px; max-width: 100%;" width="448"></iframe></div>
<div style="color: #222222;">
<b>[슬라이드]</b></div>
<div style="color: #222222;">
<br /></div>
</div>
<span style="font-size: x-small;"><span style="background-color: #fefdfa; color: #333333; font-family: "arial" , "helvetica" , sans-serif; line-height: 14px; text-align: justify;"><br /></span></span><span style="font-size: x-small;"><span style="background-color: #fefdfa; color: #333333; font-family: "arial" , "helvetica" , sans-serif; line-height: 14px; text-align: justify;">* 테리의 딥러닝 토크</span></span><br />
<span style="font-size: x-small;"><span style="background-color: #fefdfa; color: #333333; font-family: "arial" , "helvetica" , sans-serif; line-height: 14px; text-align: justify;">[youtube] <a href="https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq">https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq</a> </span></span><br />
<span style="font-size: x-small;"><span style="background-color: #fefdfa; color: #333333; font-family: "arial" , "helvetica" , sans-serif; line-height: 14px; text-align: justify;">[facebook] <a href="https://www.facebook.com/deeplearningtalk/">https://www.facebook.com/deeplearningtalk/</a></span></span><br />
<span style="font-size: x-small;"><span style="background-color: #fefdfa; color: #333333; font-family: "arial" , "helvetica" , sans-serif; line-height: 14px; text-align: justify;"><br /></span><span style="background-color: #fefdfa; color: #333333; font-family: "arial" , "helvetica" , sans-serif; line-height: 14px; text-align: justify;">* T-Robotics의 글은 facebook과 rss reader로도 받아보실 수 있습니다.</span></span><br />
<div class="separator" style="background-color: #fefdfa; clear: both; color: #333333; font-family: verdana, geneva, sans-serif; line-height: 18px; text-align: justify;">
<span style="font-family: "arial" , "helvetica" , sans-serif; font-size: x-small;">[<span class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="189cb4ff-7cae-4873-8436-d55ddbe2c9f6" id="b095b228-3a62-4807-8be7-f313dc7b6aa8"><gs class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="61cefc52-e6f8-427a-999c-d8e2bd5d8157" id="9028c2f4-078e-43ba-9375-3f608197542a"><gs class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="1db50239-7e8d-4f62-a464-d9e4af1d255e" id="6c27a77b-a00b-4dea-ad4e-58c92c705bf7">facebook</gs></gs></span><span class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="6c2e595c-3435-49db-a163-72e32681ebb2" id="a9fd5842-c391-483e-bb95-bf8f2fa72c85">]</span> <a href="http://facebook.com/trobotics" style="color: #7d181e; text-decoration-line: none;">http://facebook.com/trobotics</a></span></div>
<div style="font-family: "맑은 고딕"; margin: 0px;">
<span style="font-size: x-small;"><span style="background-color: #fefdfa; color: #333333; font-family: "verdana" , "geneva" , sans-serif; text-align: justify;">[</span><span class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="9fbdb27b-fee2-46c4-a36e-dc4f19ead275" id="1ff5bd24-5148-4c66-b842-46763eae0686" style="background-color: #fefdfa; color: #333333; font-family: "verdana" , "geneva" , sans-serif; text-align: justify;"><gs class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="df400c24-2e15-4c1e-ad29-f5e8fc8f4881" id="d7151f55-e778-4855-ab9f-bd47817d510e"><gs class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="caf4142c-4d49-46d3-a002-1f57a5fa5a22" id="13516183-5b6c-4dbf-9040-b5088aa14c77">rss</gs></gs></span><span style="background-color: #fefdfa; color: #333333; font-family: "verdana" , "geneva" , sans-serif; text-align: justify;">] </span><a href="http://t-robotics.blogspot.kr/feeds/posts/default" style="background-color: #fefdfa; color: #7d181e; font-family: arial, helvetica, sans-serif; line-height: 14px; text-align: justify; text-decoration-line: none;">http://t-robotics.blogspot.kr/feeds/posts/default</a></span><br />
<div>
<br /></div>
</div>
Terryhttp://www.blogger.com/profile/10286616328664292643noreply@blogger.com0tag:blogger.com,1999:blog-2827101888996304029.post-44258906470550907572017-04-16T07:24:00.000-07:002017-04-16T07:24:52.397-07:00#21. Deformable Convolutional NetworksTensorflow KR에서 진행하고 있는 논문읽기 모임 PR12에서 제가 deformable convolutional network에 대한 리뷰를 발표하였습니다. "고정된 형태의 convolution"을 깬 유연한 convolution으로의 진화에 대해 한번 들어보시죠. 더불어 object detection, semantic segmentation에 대한 다양한 "다음 읽을 논문"도 만나보실 수 있습니다. Spatial Transform Network (2014)를 포함해서요. <br />
<br />
"Deformable Convolutional Networks", J. Dai, H. Qi, Y. Xiong, Y. Li, G. Zhang, H. Hu, and Y. Wei, 2017.<br />
[Link] <a href="https://arxiv.org/abs/1703.06211">https://arxiv.org/abs/1703.06211</a><br />
<br />
[Slides] <a href="https://www.slideshare.net/TerryTaewoongUm/deformable-convolutional-network-2017">https://www.slideshare.net/TerryTaewoongUm/deformable-convolutional-network-2017</a><br />
<br />
<div class="wp-caption aligncenter" id="attachment_25279" style="background-color: #f3f3f3; border-radius: 1px 3px 3px; border: 1px solid rgb(221, 221, 221); color: #222222; float: none; line-height: 21px; margin: 20px auto; max-width: 100%; outline: 0px; padding: 4px 0px 0px; text-align: center; vertical-align: baseline; width: 610px;">
<br />
<iframe allowfullscreen="" frameborder="0" height="252" src="https://www.youtube.com/embed/RRwaz0fBQ0Y?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq" width="448"></iframe><br />
<b>[비디오]</b><br />
<br />
<div class="separator" style="clear: both; text-align: center;">
<br /></div>
<div class="separator" style="clear: both;">
<iframe allowfullscreen="" frameborder="0" height="380" marginheight="0" marginwidth="0" scrolling="no" src="//www.slideshare.net/slideshow/embed_code/key/QSjU3XE9twtgU" style="border-width: 1px; border: 1px solid #ccc; margin-bottom: 5px; max-width: 100%;" width="448"> </iframe> </div>
<div style="margin-bottom: 5px;">
<strong> <a href="https://www.slideshare.net/TerryTaewoongUm/deformable-convolutional-network-2017" target="_blank" title="Deformable Convolutional Network (2017)">Deformable Convolutional Network (2017)</a> </strong> from <strong><a href="https://www.slideshare.net/TerryTaewoongUm" target="_blank">University of Waterloo</a></strong> </div>
<b>[슬라이드]</b><br />
<br /></div>
<span style="font-size: x-small;"><span style="background-color: #fefdfa; color: #333333; font-family: "arial" , "helvetica" , sans-serif; line-height: 14px; text-align: justify;"><br /></span></span><span style="font-size: x-small;"><span style="background-color: #fefdfa; color: #333333; font-family: "arial" , "helvetica" , sans-serif; line-height: 14px; text-align: justify;">* 테리의 딥러닝 토크</span></span><br />
<span style="font-size: x-small;"><span style="background-color: #fefdfa; color: #333333; font-family: "arial" , "helvetica" , sans-serif; line-height: 14px; text-align: justify;">[youtube] <a href="https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq">https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq</a> </span></span><br />
<span style="font-size: x-small;"><span style="background-color: #fefdfa; color: #333333; font-family: "arial" , "helvetica" , sans-serif; line-height: 14px; text-align: justify;">[facebook] <a href="https://www.facebook.com/deeplearningtalk/">https://www.facebook.com/deeplearningtalk/</a></span></span><br />
<span style="font-size: x-small;"><span style="background-color: #fefdfa; color: #333333; font-family: "arial" , "helvetica" , sans-serif; line-height: 14px; text-align: justify;"><br /></span><span style="background-color: #fefdfa; color: #333333; font-family: "arial" , "helvetica" , sans-serif; line-height: 14px; text-align: justify;">* T-Robotics의 글은 facebook과 rss reader로도 받아보실 수 있습니다.</span></span><br />
<div class="separator" style="background-color: #fefdfa; clear: both; color: #333333; font-family: verdana, geneva, sans-serif; line-height: 18px; text-align: justify;">
<span style="font-family: "arial" , "helvetica" , sans-serif; font-size: x-small;">[<span class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="189cb4ff-7cae-4873-8436-d55ddbe2c9f6" id="b095b228-3a62-4807-8be7-f313dc7b6aa8"><gs class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="61cefc52-e6f8-427a-999c-d8e2bd5d8157" id="9028c2f4-078e-43ba-9375-3f608197542a"><gs class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="1db50239-7e8d-4f62-a464-d9e4af1d255e" id="6c27a77b-a00b-4dea-ad4e-58c92c705bf7">facebook</gs></gs></span><span class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="6c2e595c-3435-49db-a163-72e32681ebb2" id="a9fd5842-c391-483e-bb95-bf8f2fa72c85">]</span> <a href="http://facebook.com/trobotics" style="color: #7d181e; text-decoration: none;">http://facebook.com/trobotics</a></span></div>
<div style="-webkit-text-stroke-width: 0px; color: black; font-family: "맑은 고딕"; font-size: medium; font-style: normal; font-variant-caps: normal; font-variant-ligatures: normal; font-weight: normal; letter-spacing: normal; margin: 0px; orphans: 2; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; widows: 2; word-spacing: 0px;">
<span style="font-size: x-small;"><span style="background-color: #fefdfa; color: #333333; font-family: "verdana" , "geneva" , sans-serif; text-align: justify;">[</span><span class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="9fbdb27b-fee2-46c4-a36e-dc4f19ead275" id="1ff5bd24-5148-4c66-b842-46763eae0686" style="background-color: #fefdfa; color: #333333; font-family: "verdana" , "geneva" , sans-serif; text-align: justify;"><gs class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="df400c24-2e15-4c1e-ad29-f5e8fc8f4881" id="d7151f55-e778-4855-ab9f-bd47817d510e"><gs class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="caf4142c-4d49-46d3-a002-1f57a5fa5a22" id="13516183-5b6c-4dbf-9040-b5088aa14c77">rss</gs></gs></span><span style="background-color: #fefdfa; color: #333333; font-family: "verdana" , "geneva" , sans-serif; text-align: justify;">] </span><a href="http://t-robotics.blogspot.kr/feeds/posts/default" style="background-color: #fefdfa; color: #7d181e; font-family: arial, helvetica, sans-serif; line-height: 14px; text-align: justify; text-decoration: none;">http://t-robotics.blogspot.kr/feeds/posts/default</a></span></div>
<b></b><i></i><u></u><sub></sub><sup></sup><strike></strike><br />Terryhttp://www.blogger.com/profile/10286616328664292643noreply@blogger.com0tag:blogger.com,1999:blog-2827101888996304029.post-39148703496511211382017-04-10T06:54:00.000-07:002017-04-10T06:54:44.078-07:00#20. 딥러닝 핫 키워드의 변화딥러닝의 돌파구가 생긴 2012년의 AlexNet 이후, 딥러닝계에도 많은 트렌드의 변화가 있었습니다. 딥러닝 모델의 변화부터, 최적화 방법의 변화, 딥러닝 라이브러리의 변화 등등~<br />
<br />
이러한 변화들을 "키워드 언급 추이" 변화를 통해 Andrej Karpathy가 분석을 했는데요, 그 내용을 여러분께 전달해 드립니다.<br />
<br />
Andrej Karpathy의 미디엄 글 : <a href="https://medium.com/@karpathy/a-peek-at-trends-in-machine-learning-ab8a1085a106">https://medium.com/@karpathy/a-peek-at-trends-in-machine-learning-ab8a1085a106</a><br />
<br />
<div class="wp-caption aligncenter" id="attachment_25279" style="background-color: #f3f3f3; border-radius: 1px 3px 3px; border: 1px solid rgb(221, 221, 221); color: #222222; float: none; line-height: 21px; margin: 20px auto; max-width: 100%; outline: 0px; padding: 4px 0px 0px; text-align: center; vertical-align: baseline; width: 610px;">
<br />
<iframe allowfullscreen="" frameborder="0" height="252" src="https://www.youtube.com/embed/qB5Qum1FNAQ?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq" width="448"></iframe><br />
<b>[비디오]</b><br />
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEiqmO_tfBD7c5XcKtlmkn9u5QWhNMsLaNx9IiFUZoncabucg3UOw0HTASmSZ6ta4OiqHZDlZc7t4dVL6pIT-Vpc92Igual0HjB59iTjHIGo1YHXmDLj7BST9o0bpRcHqh8eg8KrLe1GS-14/s1600/Slide1.PNG" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" height="300" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEiqmO_tfBD7c5XcKtlmkn9u5QWhNMsLaNx9IiFUZoncabucg3UOw0HTASmSZ6ta4OiqHZDlZc7t4dVL6pIT-Vpc92Igual0HjB59iTjHIGo1YHXmDLj7BST9o0bpRcHqh8eg8KrLe1GS-14/s400/Slide1.PNG" width="400" /></a></div>
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEiPm8kvrPkhcz1oAS52Pyb4yFTi6aRgfGbyAK_PcqWYLdDUN0GYqtF9NsSGk2AoLpgC7qzGYsi_7QSeFhK8fKpIaF2oqm2l9PRGldBVTcZR3bpptubcItf39TiG4jFxNIupCwSXy84-hczl/s1600/Slide2.PNG" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" height="300" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEiPm8kvrPkhcz1oAS52Pyb4yFTi6aRgfGbyAK_PcqWYLdDUN0GYqtF9NsSGk2AoLpgC7qzGYsi_7QSeFhK8fKpIaF2oqm2l9PRGldBVTcZR3bpptubcItf39TiG4jFxNIupCwSXy84-hczl/s400/Slide2.PNG" width="400" /></a></div>
<div class="separator" style="clear: both;">
<b>[슬라이드]</b></div>
<br /></div>
<span style="font-size: x-small;"><span style="background-color: #fefdfa; color: #333333; font-family: "arial" , "helvetica" , sans-serif; line-height: 14px; text-align: justify;"><br /></span></span><span style="font-size: x-small;"><span style="background-color: #fefdfa; color: #333333; font-family: "arial" , "helvetica" , sans-serif; line-height: 14px; text-align: justify;">* 테리의 딥러닝 토크</span></span><br />
<span style="font-size: x-small;"><span style="background-color: #fefdfa; color: #333333; font-family: "arial" , "helvetica" , sans-serif; line-height: 14px; text-align: justify;">[youtube] <a href="https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq">https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq</a> </span></span><br />
<span style="font-size: x-small;"><span style="background-color: #fefdfa; color: #333333; font-family: "arial" , "helvetica" , sans-serif; line-height: 14px; text-align: justify;">[facebook] <a href="https://www.facebook.com/deeplearningtalk/">https://www.facebook.com/deeplearningtalk/</a></span></span><br />
<span style="font-size: x-small;"><span style="background-color: #fefdfa; color: #333333; font-family: "arial" , "helvetica" , sans-serif; line-height: 14px; text-align: justify;"><br /></span><span style="background-color: #fefdfa; color: #333333; font-family: "arial" , "helvetica" , sans-serif; line-height: 14px; text-align: justify;">* T-Robotics의 글은 facebook과 rss reader로도 받아보실 수 있습니다.</span></span><br />
<div class="separator" style="background-color: #fefdfa; clear: both; color: #333333; font-family: verdana, geneva, sans-serif; line-height: 18px; text-align: justify;">
<span style="font-family: "arial" , "helvetica" , sans-serif; font-size: x-small;">[<span class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="189cb4ff-7cae-4873-8436-d55ddbe2c9f6" id="b095b228-3a62-4807-8be7-f313dc7b6aa8"><gs class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="61cefc52-e6f8-427a-999c-d8e2bd5d8157" id="9028c2f4-078e-43ba-9375-3f608197542a"><gs class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="1db50239-7e8d-4f62-a464-d9e4af1d255e" id="6c27a77b-a00b-4dea-ad4e-58c92c705bf7">facebook</gs></gs></span><span class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="6c2e595c-3435-49db-a163-72e32681ebb2" id="a9fd5842-c391-483e-bb95-bf8f2fa72c85">]</span> <a href="http://facebook.com/trobotics" style="color: #7d181e; text-decoration: none;">http://facebook.com/trobotics</a></span></div>
<div style="-webkit-text-stroke-width: 0px; color: black; font-family: "맑은 고딕"; font-size: medium; font-style: normal; font-variant-caps: normal; font-variant-ligatures: normal; font-weight: normal; letter-spacing: normal; margin: 0px; orphans: 2; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; widows: 2; word-spacing: 0px;">
<span style="font-size: x-small;"><span style="background-color: #fefdfa; color: #333333; font-family: "verdana" , "geneva" , sans-serif; text-align: justify;">[</span><span class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="9fbdb27b-fee2-46c4-a36e-dc4f19ead275" id="1ff5bd24-5148-4c66-b842-46763eae0686" style="background-color: #fefdfa; color: #333333; font-family: "verdana" , "geneva" , sans-serif; text-align: justify;"><gs class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="df400c24-2e15-4c1e-ad29-f5e8fc8f4881" id="d7151f55-e778-4855-ab9f-bd47817d510e"><gs class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="caf4142c-4d49-46d3-a002-1f57a5fa5a22" id="13516183-5b6c-4dbf-9040-b5088aa14c77">rss</gs></gs></span><span style="background-color: #fefdfa; color: #333333; font-family: "verdana" , "geneva" , sans-serif; text-align: justify;">] </span><a href="http://t-robotics.blogspot.kr/feeds/posts/default" style="background-color: #fefdfa; color: #7d181e; font-family: arial, helvetica, sans-serif; line-height: 14px; text-align: justify; text-decoration: none;">http://t-robotics.blogspot.kr/feeds/posts/default</a></span></div>
Terryhttp://www.blogger.com/profile/10286616328664292643noreply@blogger.com0tag:blogger.com,1999:blog-2827101888996304029.post-79755057339675416092017-03-17T16:38:00.002-07:002017-03-17T16:38:29.343-07:00#19. 꼭 딥러닝을 써야하나요?많은 분들이 딥러닝을 배우고 계시고 이것을 현업에 적용하고 싶어 하십니다. 그래서 모두 건너뛰고 딥러닝만 파고 계시는 분들도 많죠.<br />
<br />
하지만 딥러닝이 꼭 다른 머신러닝들보다 좋은 성능을 내느냐 하면 그건 아닙니다. 머신러닝엔 "더 좋은" 알고리즘이란건 없고 본인의 문제에 "더 적합한" 머신러닝이 있을 뿐이니까요.<br />
<br />
"머신러닝은 프로그래밍이 아니라 디버깅이다."<br />
<br />
이 말의 의미를 한번 알아보시죠.<br />
<br />
<div class="wp-caption aligncenter" id="attachment_25279" style="background-color: #f3f3f3; border-radius: 1px 3px 3px; border: 1px solid rgb(221, 221, 221); color: #222222; float: none; line-height: 21px; margin: 20px auto; max-width: 100%; outline: 0px; padding: 4px 0px 0px; text-align: center; vertical-align: baseline; width: 610px;">
<br />
<iframe allowfullscreen="" frameborder="0" height="252" src="https://www.youtube.com/embed/MSWCQ9lRA1Y?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq" width="448"></iframe><br />
<b>[비디오]</b><br />
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEgNHRgukIktX4CWeQeLlPWSOyetpnUmvmQHw6UVqQcItJCAllE6IRx5PMOO_El6j0-Oa2z6GRywh4YhKUefXUURWYOFo0GJPf38AH4Zl2FCDkdMrw3U6fqtNJ-yVavPDWzLlw9HiAclu7bT/s1600/Slide1.PNG" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" height="300" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEgNHRgukIktX4CWeQeLlPWSOyetpnUmvmQHw6UVqQcItJCAllE6IRx5PMOO_El6j0-Oa2z6GRywh4YhKUefXUURWYOFo0GJPf38AH4Zl2FCDkdMrw3U6fqtNJ-yVavPDWzLlw9HiAclu7bT/s400/Slide1.PNG" width="400" /></a></div>
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEhvcfL5AYzADwGDqhTgLqX84MVg2DZ_ybNCj4GxoO_NFJsBoXXg-bFSUhDYTHNBXyS29MBMkHJIYO2FXj7LuUNNNOX5ePLkd5XDoB1_m0lOhf6LVy-vBYBSDwyC86R04elhJQwU7F13np-5/s1600/Slide2.PNG" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" height="300" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEhvcfL5AYzADwGDqhTgLqX84MVg2DZ_ybNCj4GxoO_NFJsBoXXg-bFSUhDYTHNBXyS29MBMkHJIYO2FXj7LuUNNNOX5ePLkd5XDoB1_m0lOhf6LVy-vBYBSDwyC86R04elhJQwU7F13np-5/s400/Slide2.PNG" width="400" /></a></div>
<div class="separator" style="clear: both; text-align: center;">
<br /></div>
<div class="separator" style="clear: both;">
<b>[슬라이드]</b></div>
<br /></div>
<span style="font-size: x-small;"><span style="background-color: #fefdfa; color: #333333; font-family: "arial" , "helvetica" , sans-serif; line-height: 14px; text-align: justify;"><br /></span></span><span style="font-size: x-small;"><span style="background-color: #fefdfa; color: #333333; font-family: "arial" , "helvetica" , sans-serif; line-height: 14px; text-align: justify;">* 테리의 딥러닝 토크</span></span><br />
<span style="font-size: x-small;"><span style="background-color: #fefdfa; color: #333333; font-family: "arial" , "helvetica" , sans-serif; line-height: 14px; text-align: justify;">[youtube] <a href="https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq">https://www.youtube.com/playlist?list=PL0oFI08O71gKEXITQ7OG2SCCXkrtid7Fq</a> </span></span><br />
<span style="font-size: x-small;"><span style="background-color: #fefdfa; color: #333333; font-family: "arial" , "helvetica" , sans-serif; line-height: 14px; text-align: justify;">[facebook] <a href="https://www.facebook.com/deeplearningtalk/">https://www.facebook.com/deeplearningtalk/</a></span></span><br />
<span style="font-size: x-small;"><span style="background-color: #fefdfa; color: #333333; font-family: "arial" , "helvetica" , sans-serif; line-height: 14px; text-align: justify;"><br /></span><span style="background-color: #fefdfa; color: #333333; font-family: "arial" , "helvetica" , sans-serif; line-height: 14px; text-align: justify;">* T-Robotics의 글은 facebook과 rss reader로도 받아보실 수 있습니다.</span></span><br />
<div class="separator" style="background-color: #fefdfa; clear: both; color: #333333; font-family: verdana, geneva, sans-serif; line-height: 18px; text-align: justify;">
<span style="font-family: "arial" , "helvetica" , sans-serif; font-size: x-small;">[<span class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="189cb4ff-7cae-4873-8436-d55ddbe2c9f6" id="b095b228-3a62-4807-8be7-f313dc7b6aa8"><gs class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="61cefc52-e6f8-427a-999c-d8e2bd5d8157" id="9028c2f4-078e-43ba-9375-3f608197542a"><gs class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="1db50239-7e8d-4f62-a464-d9e4af1d255e" id="6c27a77b-a00b-4dea-ad4e-58c92c705bf7">facebook</gs></gs></span><span class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="6c2e595c-3435-49db-a163-72e32681ebb2" id="a9fd5842-c391-483e-bb95-bf8f2fa72c85">]</span> <a href="http://facebook.com/trobotics" style="color: #7d181e; text-decoration: none;">http://facebook.com/trobotics</a></span></div>
<div style="-webkit-text-stroke-width: 0px; color: black; font-family: "맑은 고딕"; font-size: medium; font-style: normal; font-variant-caps: normal; font-variant-ligatures: normal; font-weight: normal; letter-spacing: normal; margin: 0px; orphans: 2; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; widows: 2; word-spacing: 0px;">
<span style="font-size: x-small;"><span style="background-color: #fefdfa; color: #333333; font-family: "verdana" , "geneva" , sans-serif; text-align: justify;">[</span><span class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="9fbdb27b-fee2-46c4-a36e-dc4f19ead275" id="1ff5bd24-5148-4c66-b842-46763eae0686" style="background-color: #fefdfa; color: #333333; font-family: "verdana" , "geneva" , sans-serif; text-align: justify;"><gs class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="df400c24-2e15-4c1e-ad29-f5e8fc8f4881" id="d7151f55-e778-4855-ab9f-bd47817d510e"><gs class="GINGER_SOFTWARE_mark" ginger_software_uiphraseguid="caf4142c-4d49-46d3-a002-1f57a5fa5a22" id="13516183-5b6c-4dbf-9040-b5088aa14c77">rss</gs></gs></span><span style="background-color: #fefdfa; color: #333333; font-family: "verdana" , "geneva" , sans-serif; text-align: justify;">] </span><a href="http://t-robotics.blogspot.kr/feeds/posts/default" style="background-color: #fefdfa; color: #7d181e; font-family: arial, helvetica, sans-serif; line-height: 14px; text-align: justify; text-decoration: none;">http://t-robotics.blogspot.kr/feeds/posts/default</a></span></div>
Terryhttp://www.blogger.com/profile/10286616328664292643noreply@blogger.com0