Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

怖くない誤差逆伝播法 Chainerを添えて

Introduction of backpropagation with chainer

  • Login to see the comments

怖くない誤差逆伝播法 Chainerを添えて

  1. 1. Komachi Lab M2 2016/06/14 Chainer back propagation
  2. 2. Komachi Lab 2
  3. 3. Komachi Lab Neural Net 1. forward 2. Loss Loss 3. 4. (AdaGrad ) 5. 1 3 SGD
  4. 4. Komachi Lab Neural Net 2. Loss Loss 3. 4 W Loss Loss
  5. 5. Komachi Lab Neural Net 1. forward 2. Loss Loss 3. 4. (AdaGrad ) 5. 1 5 SGD W1 W2 predict input
  6. 6. Komachi Lab Neural Net 1. forward 2. Loss Loss 3. 4. (AdaGrad ) 5. 1 6 SGD Loss func correct W1 W2 predict input
  7. 7. Komachi Lab Neural Net 1. forward 2. Loss Loss 3. 4. (AdaGrad ) 5. 1 7 SGD gW1 gW2 Loss func correct W1 W2 predict input
  8. 8. Komachi Lab Neural Net 1. forward 2. Loss Loss 3. 4. (AdaGrad ) 5. 1 8 SGD gW1 gW2 Loss func correct W1 W2 predict input
  9. 9. Komachi Lab 9
  10. 10. Komachi Lab 1. forward 2. Loss Loss 3. 4. (AdaGrad ) 5. 1 10 gW1 gW2 Loss func correct W1 W2 predict input
  11. 11. Komachi Lab ✤ 11
  12. 12. Komachi Lab 12
  13. 13. Komachi Lab ✤ W1 W2 ✤ x, h, y, t ✤ Loss 13 (i train set index) correct W1 W2 predict input :x hidden:h :y :t
  14. 14. Komachi Lab ✤ 14 W1, W2 correct W1 W2 predict input :x hidden:h :y :t
  15. 15. Komachi Lab ✤ 
 i 15 forward correct W1 W2 predict input :x hidden:h :y :t
  16. 16. Komachi Lab 16 gW2 gW1
  17. 17. Komachi Lab 17 z W4 x W3 h W1 y W2
  18. 18. Komachi Lab 18 (Chainer )
  19. 19. Komachi Lab 19
  20. 20. Komachi Lab on Chainer 20
  21. 21. Komachi Lab Chainer ✤ PFN DNN 
 http://chainer.org/ ✤ Define-by-Run 
 Define-and-Run ✤ 
 TensorFlow, Theano, Torch, Keras, Caffe, etc... 21
  22. 22. Komachi Lab on Chainer 22
  23. 23. Komachi Lab 23 loss.backward()
  24. 24. Komachi Lab 24 loss.backward() loss backward
  25. 25. Komachi Lab MeanSquaredError 25 x0, x1 y, t backward gy 1 backward forward inputs gy backward gy
  26. 26. Komachi Lab Linear 26 grad_outputs Mean Squared Error backward gW backward
  27. 27. Komachi Lab Linear 27 grad_outputs Linear backward
  28. 28. Komachi Lab on Chainer ✤ ✤ ✤ ✤ W 28 tanh forward backward chainer.functions.activation.tanh forward Chain backward Chain Chainer
  29. 29. Komachi Lab ✤ - ✤ Chainer - - 29
  30. 30. Komachi Lab LSTM 30 forward a, i, f, o Linear backward gy

×