[B! deeplearning] chezouã®ãƒ–ãƒƒã‚¯ãƒžãƒ¼ã‚¯

chezou id:chezou

deeplearningã«é–¢ã™ã‚‹chezouã®ãƒ–ãƒƒã‚¯ãƒžãƒ¼ã‚¯ (11)

${{author_name}}$

{{{comment_expanded}}}

{{label}}

{{#is_bookmark}}ãƒªã‚¹ãƒˆ{{/is_bookmark}}{{^is_bookmark}}ãƒªãƒ³ã‚¯{{/is_bookmark}}

${{author_name}}$
{{author_name}}{{created}}
{{ #comment }}{{ comment }}{{ /comment }}
- {{ label }}

${{author_name}}$

{{{comment_expanded}}}

{{label}}

{{#is_bookmark}}ãƒªã‚¹ãƒˆ{{/is_bookmark}}{{^is_bookmark}}ãƒªãƒ³ã‚¯{{/is_bookmark}}

æ·±å±¤å¦ç¿’ã«ã‚ˆã‚‹è‡ªç„¶è¨€èªžå‡¦ç† - RNN, LSTM, ãƒ‹ãƒ¥ãƒ¼ãƒ©ãƒ«æ©Ÿæ¢°ç¿»è¨³ã®ç†è«–
æœ¬ç¨¿ã§ã¯ãƒ‹ãƒ¥ãƒ¼ãƒ©ãƒ«ãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯ï¼Œèª¤å·®é€†ä¼æ’æ³•ï¼Œè¨€èªžãƒ¢ãƒ‡ãƒ«ï¼ŒRNNï¼ŒLSTMï¼Œãƒ‹ãƒ¥ãƒ¼ãƒ©ãƒ«æ©Ÿæ¢°ç¿»è¨³ã®ä¸€é€£ã®æ‰‹æ³•ã«ã¤ã„ã¦æ•°ç†çš„ã«è§£èª¬ã™ã‚‹ï¼Ž å‰ç·¨ã®ç›®æ¬¡ ãƒ‹ãƒ¥ãƒ¼ãƒ©ãƒ«ãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯ é †ä¼æ’ (Forwardpropagation) é€†ä¼æ’ (Backpropagation) ãƒªã‚«ãƒ¬ãƒ³ãƒˆãƒ‹ãƒ¥ãƒ¼ãƒ©ãƒ«ãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯ (RNN) Recurrent Neural Network Language Model (RNNLM) Backpropagation Through Time (BPTT) Long Short-Term Memory (LSTM) Gated Recurrent Unit (GRU) RNN ã®ãƒ‰ãƒãƒƒãƒ—ã‚¢ã‚¦ãƒˆã¨ãƒãƒƒãƒæ£è¦åŒ– ãƒ‹ãƒ¥ãƒ¼ãƒ©ãƒ«æ©Ÿæ¢°ç¿»è¨³ (NMT) Sequence to Sequence (seq2seq) æ³¨æ„ (Attention) åŒæ–¹å‘ã‚¨ãƒ³ã‚³ãƒ¼ãƒ€ãƒ¼ãƒ»å¤šå±¤LSTM è©•ä¾¡æ‰‹æ³•
chezou 2017/03/02
deeplearning
ãƒªãƒ³ã‚¯
è‡ªå·±ç¬¦å·åŒ–å™¨ã‚’ç”¨ã„ãŸDCGANã®äº‹å‰å¦ç¿’ - ãƒ‹ãƒ¼ãƒˆãŒãƒ—ãƒã‚°ãƒ©ãƒŸãƒ³ã‚°ã™ã‚‹ãƒ–ãƒã‚°ï¼ˆã¯ã¦ãªå‡ºå¼µæ‰€ï¼‰
2016 - 12 - 01 è‡ªå·±ç¬¦å·åŒ–å™¨ã‚’ç”¨ã„ãŸDCGANã®äº‹å‰å¦ç¿’ æ³¨æ„ï¼šã¾ã ç ”ç©¶ä¸ãªã®ã§ã“ã®ãƒšãƒ¼ã‚¸ã®å†…å®¹ã¯é–“é•ã£ã¦ã„ãŸã‚Šã—ã¦ã‚‹ã¨æ€ã„ã¾ã™ã€‚ ä»Šã•ã‚‰ã§ã™ãŒDCGANã«æ‰‹ã‚’å‡ºã—ã¦ã¿ã¾ã—ãŸã€‚ ãŸã è©¦ã™ã ã‘ã§ã¯å‘³æ°—ãªã„ã®ã§å°‘ã—ã‚¢ãƒ¬ãƒ³ã‚¸ã—ã¦ã¿ã¾ã—ãŸã€‚ ãƒ‡ãƒ¼ã‚¿ã¯sugyanã•ã‚“ã®ã‚¢ã‚¤ãƒ‰ãƒ«ç”»åƒãƒ‡ãƒ¼ ã‚¿ã‚»ãƒƒãƒˆ ã‚’ä½¿ã„ã¾ã—ãŸã€‚ 1.å‰ç½®ã æ˜¨å¹´ã®ä»Šé ƒã«DCGAN(Deep Convolutional Generative Adversarial Networks)ãŒè©±é¡Œã¨ãªã£ã¦ã„ã¾ã—ãŸã€‚ DCGANã¨ã¯ã€ä¸€æ§˜ä¹±æ•°ã®100æ¬¡å…ƒãƒ™ã‚¯ãƒˆãƒ«ã‹ã‚‰ãã‚Œã„ãªç”»åƒã‚’ç”Ÿæˆã™ã‚‹generatorã‚’ä½œã‚‹ã€ã¨ã„ã†ã‚‚ã®ã§ã™ã€‚ ä¾‹ãˆã°DCGANã‚’ä½¿ãˆã°ã€ ã‚¢ãƒ‹ãƒ¡é¡”ã‚’ç”Ÿæˆ ã—ãŸã‚Šã€ ã‚¢ã‚¤ãƒ‰ãƒ«ã®é¡”ã‚’ç”Ÿæˆ ã—ãŸã‚Šã§ãã¾ã™ã€‚ DCGANã®è©³ã—ã„å†…å®¹ã¯ ã“ã¡ã‚‰ã®ãƒšãƒ¼ã‚¸ ã‚’ã”è¦§ãã ã•ã„ã€‚ ã¨ã“ã‚ã§ã€ ã“ã¡ã‚‰ã®ãƒšãƒ¼ã‚¸ ã«ã‚ˆã‚Œã°ã€DCGANã¯å¦
chezou 2016/12/01
deeplearning
ãƒªãƒ³ã‚¯
DNC (Differentiable Neural Computers) ã®æ¦‚è¦ + Chainer ã«ã‚ˆã‚‹å®Ÿè£… - Qiita
ã“ã‚“ã«ã¡ã¯ï¼Ž@yos1up ã¨ç”³ã—ã¾ã™ï¼Žå¥½ããªé£Ÿã¹ç‰©ã¯ï¼Œç·‘è‰²ã®ã‚ãƒŽã‚³ã§ã™ï¼Ž 2016/10/12ã« DeepMind ãŒ Nature ã«æŠ•ç¨¿ã—ãŸè«–æ–‡ï¼ŒHybrid computing using a neural network with dynamic external memory ã«ã¦ææ¡ˆã•ã‚Œã¦ã„ã‚‹ãƒ‹ãƒ¥ãƒ¼ãƒ©ãƒ«ãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯ãƒ¢ãƒ‡ãƒ« DNC (Differentiable Neural Computers) ã‚’ï¼Œå¤§æ€¥ãŽã§ Chainer ã§å®Ÿè£…ã—ã¦ã¿ã¾ã—ãŸï¼Ž DNC ã«ã¤ã„ã¦ DNC ã¯ï¼Œå…ˆè¿°ã®è«–æ–‡ã§ææ¡ˆã•ã‚ŒãŸæ–°ã—ã„ãƒ‹ãƒ¥ãƒ¼ãƒ©ãƒ«ãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯ã§ï¼Œãã®æƒ…å ±å‡¦ç†èƒ½åŠ›ã®é«˜ã•ãŒæœŸå¾…ã•ã‚Œã¦ã„ã¾ã™ï¼Žè«–æ–‡ä¸ã§ã¯ï¼Œã‚°ãƒ©ãƒ•ä¸Šã®æœ€çŸçµŒè·¯ã‚¿ã‚¹ã‚¯ã‚„ã¡ã‚‡ã£ã¨ã—ãŸãƒ‘ã‚ºãƒ«ã®ã‚¿ã‚¹ã‚¯ãªã©ï¼Œå¾“æ¥ãƒ‹ãƒ¥ãƒ¼ãƒ©ãƒ«ãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯ã§ã¯å¦ç¿’ä¸å¯èƒ½ã¨æ€ã‚ã‚Œã¦ã„ãŸã‚¿ã‚¹ã‚¯ãŒ DNC ã§å¦ç¿’ã§ãã¦ãŠã‚Šï¼Œãã®æƒ…å ±å‡¦ç†èƒ½åŠ›ã®é«˜ã•ãŒã†ã‹ãŒã„çŸ¥ã‚Œã¾ã™ï¼Ž DN
chezou 2016/10/21
DeepLearning
ãƒªãƒ³ã‚¯
ç¾åœ¨ã®DNNã«ãŠã‘ã‚‹æœªè§£æ±ºå•é¡Œ
è„³åž‹è¨ˆç®—æ©Ÿé›‘è«‡ä¼šã§ã®è³‡æ–™ã§ã™ 1. å¤§ããªNNã®å¦ç¿’ã¯ãªãœä¸€æ§˜ã«æˆåŠŸã™ã‚‹ã‹ 2. æ•µå¯¾çš„ç”Ÿæˆãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯(GAN)ã®è§£æž 3. seq2seqã«ã‚ˆã‚‹å¯å¤‰é•·æƒ…å ±ã®åŸ‹ã‚è¾¼ã¿ 4. Ladder Networkã®è§£æž
chezou 2016/05/26
deepLearning
ãƒªãƒ³ã‚¯
Implementing Recurrent Neural Net using chainer!
I just started to study deep learning, which is huge boom both in academia and industry. A week ago, a Japanese company called Â Preferred Infrastructure (PFI) released new deep learning framework " chainer"! This framework is really great. I was able to implement a recurrent net with less than 100 lines of python code. Specifically, I tried new recurrent neural network (RNN) called IRNN described
chezou 2015/06/16
Chainerã®ç·´ç¿’ãŒã¦ã‚‰æœ€æ–°ã®Hintonã®è«–æ–‡ã®IRNNã‚’å®Ÿè£…ã—ãŸã‚‰100è¡Œè¡Œã‹ãªã‹ã£ãŸðŸ˜›ã£ã¦è¨˜äº‹ã

deeplearning

chainer
ãƒªãƒ³ã‚¯
ãã‚“ã„ã‚DeepLearning - White scenery @showyou, hatena
http://deep-learning-hackathon.connpass.com/event/12867/ ã“ã®ã‚¤ãƒ™ãƒ³ãƒˆã«å‚åŠ ã—ã¦ã€DeepLearning(Caffe)ã‚’å›žã—ã¦ã¿ã¾ã—ãŸã€‚ å¤§ä½“ã“ã‚Œã¾ã§ä¸ŠãŒã£ã¦ã‚‹è¨˜äº‹ã§ã€ã‚´ãƒã†ã•ã§DeepLearningã•ã‚Œã¦ã‚‹ã®ãŒã‚ã£ãŸã®ã§å¯¾æŠ—ã—ã¦ãã‚“ãƒ¢ã‚¶ã§ã‚„ã£ã¦ã¿ã¾ã—ãŸã€‚ï½Žç•ªç…Žã˜ã§ã™ã€‚ï¼ˆï¼ˆæ—¢ã«ãƒ©ãƒ–ãƒ©ã‚¤ãƒ–ç‰ã§ã‚‚ã•ã‚Œã¦ã‚‹ã‚ˆã†ãªã®ã§ï¼‰ï¼‰ ç’°å¢ƒ Dynabook RX3(Core i5 M520, 4GBMem, Linux Mint 17.1 64bit) Amazon EC2 g2.2xlarge ã‚¤ãƒ³ã‚¹ã‚¿ãƒ³ã‚¹ å•é¡Œè¨å®š äº‹å‰ã«animefaceã§é¡”é ˜åŸŸã‚’æ¤œå‡ºã—ãŸéƒ¨åˆ†ã«å¯¾ã—ã€ã‚¢ãƒªã‚¹ã§ã‚ã‚‹ã‹ã‚«ãƒ¬ãƒ³ã§ã‚ã‚‹ã‹ã€ãã‚Œã¨ã‚‚ä»–ã®ã‚ãƒ£ãƒ©ã§ã‚ã‚‹ã‹ã‚’åˆ¤å®šã—ãŸã„ å˜ç´”ãªï¼“ã‚¯ãƒ©ã‚¹åˆ†é¡žå•é¡Œã§ã™ãŒã€ã‚¢ãƒªã‚¹ã‚‚ã‚«ãƒ¬ãƒ³ã‚‚ã‚ãƒ³ãƒ‘ãƒ„ãªã®ã§ã€è‰²ã«ã‚ˆã‚‹ã‚ãƒ£ãƒ©åˆ¤åˆ¥ã¯ã§ãã¾ã›ã‚“ã€‚ãŠãã‚‰ãè¼ªéƒã‚„
chezou 2015/05/25
deeplearning
ãƒªãƒ³ã‚¯
ã‚¨ãƒ¬ã‚¯ãƒˆãƒãƒ»ã‚·ã‚¹ãƒ†ãƒ æ ªå¼ä¼šç¤¾ å…¬å¼ Google+
chezou 2015/05/06
DQNã£ã¦è©¦ã›ã‚‹å®Ÿè£…ã‚ã£ãŸã‚“ã

deeplearning
ãƒªãƒ³ã‚¯
Images to Text - Toronto Deep Learning
Due to heavy traffic, server might be slow... Also try the Static Demo.
chezou 2015/04/21
Crowd Sourcingã§ã‚ãƒ£ãƒ—ã‚·ãƒ§ãƒ³ã‚’ã¤ã‘ãŸã®ã‚’å¦ç¿’ã—ãŸã¨ã®ã“ã¨ #rtechlab

deeplearning

nlp
ãƒªãƒ³ã‚¯
ãƒ‡ã‚£ãƒ¼ãƒ—ãƒ©ãƒ¼ãƒ‹ãƒ³ã‚°ãƒãƒ¥ãƒ¼ãƒˆãƒªã‚¢ãƒ« å¿œç”¨ç·¨
Transcript 1. å¤§è¦æ¨¡ãƒ‡ãƒ¼ã‚¿ã‹ã‚‰å˜èªžã® æ„å‘³è¡¨ç¾å¦ç¿’-word2vec ãƒœãƒ¬ã‚¬ãƒ©â€ƒãƒ€ãƒŒã‚·ã‚«â€ƒåšå£«(æƒ…å ±ç†å·¥å¦) è‹±å›½ãƒªãƒãƒ¼ãƒ—ãƒ¼ãƒ«å¤§å¦è¨ˆç®—æ©Ÿç§‘å¦ç§‘å‡†æ•™æŽˆ 2. 2 2005 2008~10 å¦éƒ¨ ä¿®å£« åšå£« åŠ©æ•™/è¬›å¸« æ±äº¬å¤§å¦ å·¥å¦éƒ¨ æ±äº¬å¤§å¦å¤§å¦é™¢æƒ…å ±ç†å·¥å¦ç³» æ–‡æ›¸è‡ªå‹•è¦ç´„ã«ãŠã‘ã‚‹ é‡è¦æ–‡é †åºå¦ç¿’ åŒå§“åŒåæŠ½å‡º åˆ¥åæŠ½å‡º å±žæ€§é¡žä¼¼æ€§è¨ˆæ¸¬ é–¢ä¿‚é¡žä¼¼æ€§è¨ˆæ¸¬ è©•åˆ¤åˆ†é¡žã®åˆ†é‡Žé©å¿œ é–¢ä¿‚æŠ½å‡ºã®åˆ†é‡Žé©å¿œ é€²åŒ–è¨ˆç®—ã‚’ç”¨ã„ãŸWeb æ¤œç´¢çµæžœé †åºå¦ç¿’ ã‚½ãƒ¼ã‚·ãƒ£ãƒ«ãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯ ã®é–¢ä¿‚äºˆæ¸¬ å¯¾è©±åž‹å”èª¿ Webæ¤œç´¢ã‚¨ãƒ³ã‚¸ãƒ³ æ½œåœ¨é–¢ä¿‚æ¤œç´¢ ã‚¨ãƒ³ã‚¸ãƒ³ è‡ªå·±ç´¹ä»‹ å°‚é–€åˆ†é‡Žï¼šè‡ªç„¶è¨€èªžå‡¦ç†, æ©Ÿæ¢°å¦ç¿’,ãƒ‡ãƒ¼ã‚¿ãƒžã‚¤ãƒ‹ãƒ³ã‚° 2006~07 2010~13 2010~ç¾åœ¨ å‡†æ•™æŽˆ ãƒªãƒãƒ¼ãƒ—ãƒ¼ãƒ«å¤§å¦ æ·±å±¤å¦ç¿’ 3. ä»Šå›žã®è¬›æ¼”ã®èƒŒæ™¯ â€¢æ·±å±¤å¦ç¿’ã«é–¢ã™ã‚‹æ´»å‹• â€¢2014å¹´9æœˆã«æ·±å±¤å¦ç¿’ã®ãƒãƒ¥ãƒ¼ãƒˆãƒªã‚¢ãƒ«ã‚’CyberAge
chezou 2015/04/17
deeplearning
ãƒªãƒ³ã‚¯
Portfolio
Introduction Approximately 25 years of experience Passionate about data privacy and censorship resistance Lead a team of ~30 engineers from hiring to delivering the product to the first customer. Worked in following projects / companies Ethereum Foundation ( worked on Swarm when it was part of Ethereum geth code) Swarm Foundation Prysmatic Labs (worked on PoS Merge) Shardeum Foundation (worked on
chezou 2014/12/30
deeplearning
ãƒªãƒ³ã‚¯
Deep Learning ã‚’å®Ÿè£…ã™ã‚‹
Building High-level Features Using Large Scale Unsupervised Learning
chezou 2013/02/06
deeplearning

Theano
ãƒªãƒ³ã‚¯
1