2017 Deep Learning and Psychology Gakusyuin 01

学習院大学心理学特殊講義深層学習の心理学的解釈第1回資料

Transcript

1. 1. : : ( ) asakawa@ieee.org : 2 : 2-4 : 1 : : 5 (16:20-17:50) 16:20 #2017dl_gak 2. 3. 4. 5. 6. 7. Q and A 8. 9. 10. 11. 12. 3 13. 14. 15. 3.
2. 1: USCD : ( ) Python ( ,2016) ( ,2015) (2001) 4. ( ) 2012 ( ) 1. Caffe, 2. Theano, 3. Keras, 4. TensorFlow, 5. Chainer, 6. Touch, 7. DeepLearing4j 8. CNTK, 1. CNN 2. RNN 3. RL
3. 3 (CNN) (ReLU), (dropout), (maxout) (RNN), (DeepFace), (Network in Network), (Very Deep Convolutional Neural Network), (Sequence to sequence learning) 5. 1. 1. 2. 2. 1. 2. 3. 1. 2. 4. 1. 2. 3. 4. 5. 5. CNN 6. 1. LeNet5 2. AlexNet 3. ZFnet 4. GoogLeNet 5. VGG 6. VeryDeep 7. ResNet 8. GAN 7. 1. 2. 8. 1. Skip-gram 2. CBOW 3. 9. (1) 1. 2.
4. 3. 10. (2) 1. LSTM 2. GRU 3. 4. 5. 11. 1. 2. 3. 4. 12. 1. 2. 3. 13. 1. 2. Q 14. 1. AI 2. DQN 15. 6 4 28 iclr 5 26 jsai 7. Q and A 1. AI 2. 3. 4. 5.
5. 6. 7. http://www.jiji.com/jc/article?k=000000251.000002302&g=prt 8. 8. : WiFi PC 9. : 30 : 20 : 25 : 15 : 10 : A=90-100, B=80-89, C=70-79, D=65-69, F=0-64 9. : 30% 30 : 20 20 : 25 50 : 15 15 : 10 15 (ReLU) : A=90-100, B=80-89, C=70-79, D=65-69, F=0-64
6. 10. Python 2016 ISBN:4339028517 2014 ISBN:4788514222 2 2 11. :BC552-BC479 93 2: (BC469 -BC399)
7. 3: (1623-1662) 11.1 (384B.C-322B.C) Konwing yourself is the beginning of all wisdom. 4: http://www.biography.com/people/aristotle-9188415 11.2 (1596-1650) 5: https://en.wikipedia.org/wiki/Ren%C3%A9_Descartes
8. 6: https://en.wikipedia.org/wiki/Cartesian_theater 11.3 (1912–1954) 7: http://www.azquotes.com/author/14856-Alan_Turing 11.4 (1903-1957) 8: http://www.azquotes.com/quote/893018
9. 12. 3 9: https://leonardoaraujosantos.gitbooks.io/artificial-inteligence/content/Images/AI_DeepLearning.jpg 13. 13 (2014) ( 1)
10. artificial intelligence AI SF ( 2) 2001 HAL9000 2000 R2D2 ( 3) GPX AI ( 4) ( 5) 1950 1956 (Dartmouth Conference)
11. J. McCarthy Artificial Intelligence 1945 ENIAC 1950 (A. M. Turing) “Computer Machinery and Intelligence” 1950 (C. Shannon) “Automatic Chess Player” 1958 (J. MacCarthy) LISP 1956 (1980 )( 6) 1950-1960 1969 (P. J. Hayes)”Some Philosophical Problems from the Standpoint of Artificial Intelligence” => GOFAI(Good Old Fashioned AI: AI) 1970 1972 (T. Winograd) “Natural Language Understanding” SHRDLU 1975 ”A Framework for Representing Knwoledge” ( 7) 10:
12. 1980 ( 8) 1980 1990 WWW 2000 ( 9) (1)( 10) 1 1
13. 1 11: HP http://www.ai-gakkai.or.jp/jsai/whatsai/AItopics1.html (2)( 11) 2 2 ‥‥‥‥ } (3)( 12)
14. 12: http://www.ai-gakkai.or.jp/jsai/whatsai/AItopics1.html ( 13) J. Searle Harnad (1990)( 14) 13: image ( ) 3 1. 2. Marr's Tri-Level Hypothesis computational level:
15. algorithmic/representational level: ( ) ( ) implementational/physical level: ( ) 14. 14.1 Arthur Samuel (1959) : "field of study that gives computers the ability to learn without being explicitly programmed" 14: Sir Authur Sammual 14.2 Tom Mitchell (1999) : T E P P E A computer program is said to learn from experience E with respect to some task T and some performance measure P, if its performance on T, as measured by P, improves with experience E.
16. 15: Tom Mitchell 14.3 CNN RNN 16: scikit-learn
17. 14.4 1. linear regression 2. logistic regression 3. regularization 4. multi-layered perceptrons 5. convolutional neural networks 6. recurrent neural networks 7. reinforcement learning 14.5. 1. 2. 3. 4. 5. (C, C++, Python, java, ruby, shell scripts, javascript, Haskell, Scala, ...) 6. markdown, jeykll, GitHub, 7. Linux (Ubuntu MacOS ) 8. GPU (NVIDIA ) 9. GURU 14.6. AI 5 1. ( GPUs, ASICs), 2. (e.g. ImageNet), 3. (e.g. , CNN, LSTM), and 4. (Linux, TCP/IP, Git, ROS, PR2, AWS, AMT, TensorFlow, etc.). 5. (arXiv.org) 14.7 ...
18. 17: https://xkcd.com/882/
19. 14.8 (ASA) P 1. p 2. p 3. p 4. 5. p 6. p ASA Statement on Statistical Significance and P-values 15. 15.1 14.1.1 1950 : ( ) 18: 19: ()
20. 20: 15.1.2 Rosenblatt 21:
21. 22: 23: wikipedia 15.1.3 perceptron 3 S(sensory layer), A(associative layer), R(response layer) , ( ) ( ) ( ) ( )
22. 1960 15.2 1986 PDP 1989 Authors: J.A. Anderson, A. Pellionisz, E. Rosenfeld (eds.) Title: Neurocomputing 2: Directions for Research Reference: MIT Press, Cambridge (1990), Massachusetts ANNs are some kind of non-linear statistics for amateurs 15.3 3 24: 2013 ICLR arXiv.org 2013 Mikolov word2vec
23. 25: Mikolov 2013 DeepMind DQN 0:09 / 1:08 0:08 / 1:12
24. 26: DQN 2014 Neural Image Captioning 27: Human: A group of men playing Frisbee in the park.
25. Machine: A group of young people playing a game of Frisbee. 28: Vinyals et. al (2014) 2015
26. 29: ( ) 2015 2016 GAN 30: Generative Adversarial Text to Image Synthesis arXiv:1605.05396v2 31: Generative Adversarial Text to Image Synthesis arXiv:1605.05396v2 2016
27. 2016 32: Nature 16. 1. Python Theano, Keras, TensorFlow, Chainer, CNTK, 2. MATLAB (octave) 3. Java DeepLearing4j 4. C, C++ Caffe, 5. Torch Touch, 6. FORTRAN (BLAS) Excel R(H20) frameworks Python/Java C, C++,
28. 33: https://xkcd.com/1425/ TensorFlow playground http://playground.tensorflow.org/ Convnet.js http://cs.stanford.edu/people/karpathy/convnetjs/ https://www.captioinbot.ai Alex graves, http://www.cs.toronto.edu/~graves/handwriting.html] Ink Poster: Handwritten post-it notes http://www.inkposter.com/? http://memo.sugyan.com/entry/2016/11/28/131952 prisma http://touchlab.jp/2016/07/prisma_app_review/