Haiku Generation Based on Motif Images Using Deep Learning Koki Yoneda 1 Soichiro Yokoyama 2 Tomohisa Yamashita 2 Hidenori Kawamura Scho

Size: px

Start display at page:

Download "Haiku Generation Based on Motif Images Using Deep Learning Koki Yoneda 1 Soichiro Yokoyama 2 Tomohisa Yamashita 2 Hidenori Kawamura Scho"

れいがひがき
7 years ago
Views:

1 Haiku Generation Based on Motif Images Using Deep Learning Koki Yoneda 1 Soichiro Yokoyama 2 Tomohisa Yamashita 2 Hidenori Kawamura School of Engineering Hokkaido University 2 2 Graduate School of Information Science and Technology, Hokkaido University Abstract:,.,,,, LSTM,,,.,LSTM 1 1.1,,.,Google Deep Convolution Neural Network (CNN) Deep CNN, Recurrent Neural Network (RNN)[3] RNN Google [4]., DCGAN, [5],,,17,, 1 [1][2], [email protected],,,,,,, [6][8], [7] 1.2,,,,,,,,

, 3 3 1: LSTM, 4,,, 4 3 1 tanh 1 x t f t = σ(w f [h t 1, x t ] + b f ) i t = σ(w i [h t 1, x t

2 2, tanh, o t o t = σ(w o [h t 1, x t ] + b o ), h t h t = o t tanh(c t )., 2.1, (Long Short-Term Memory, LSTM ) LSTM,, 1, (Convolutional Neurak Network, CNN ),, CNN,,,., 3 3 1: LSTM, 4,,, tanh 1 x t f t = σ(w f [h t 1, x t ] + b f ) i t = σ(w i [h t 1, x t ]+b i ) tanh C C t = tanh(w C [h t 1, x t ] + b C ) i t C,C t = f t C t 1 +i t C t 3.1,, 3, web [9] [10] [11] 17, ( 38,506

3 3.2 web [12],,, 8, , imagenavi, 1,,,,, 369,754 4, 3 ID,, BPTT ID ID, 1-of-K projection layer [14] LSTM LSTM, ID 1 LSTM, LSTM LSTM TensorFlow[15] LSTM :3 LSTM :1024 :Adam[16] :0.02 :0.99 :300 :50 :100 : 2: LSTM 4.1 3, LSTM[13] 2, 3,, ID, 3: LSTM

4 4.2, : LSTM LSTM, 5,,, 6 5: LSTM 1,,,, 4.2.1,,, 3,,,,, 17, , 1, : 5,7,5,,. 1,

5 4.3,,, 7, ID,3,. Inception-v3[17], ID 2,,,. 8: :1024,512,256 :Adam : :1000 : 9 9: 5 7: 3,LSTM, 5.1,,Levenshtein [18] Levenshtein 2, 3 10,000, Levenshtein

6 LSTM :2, 3 LSTM :256, 512, , 11 x Levenshtein,y,l,u 10 1,024, Levenshtein,. 11, Levenshtein,, 12,13,14 x, y 12 timeout,, 1 14 cannot read, ,,,5.1,, 10: 12: : 11: 5.2, LSTM : :

7 16: ( : ) 14: : 5.3,,,., 15,16,17 15, 16,17,, 15: ( : ) 6 LSTM LSTM, 17: ( : ),,LSTM,,,,,,,,, AI, NHK!,,,,,,,

8 [1] [2] 2014 [3] Shujie Liu,Nan Yang,Mu Li,Ming Zhou. A Recursive Recurrent Neural Network for Statistical Machine Translation,2014,ACL. [4] Yonghui Wu, Mike Schuster, Zhifeng Chen, et al., Google s Neural Machine Translation System: Bridging the Gap between Human and Machine Translation,2016,arXiv: [5] Alec Radford, Luke Metz, Soumith Chintala, Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks,2015,arXiv. [6] YANG, Ming, and Masafumi HAGIWARA. A Text-based Automatic Waka Generation System using Kansei. International Journal of Affective Engineering 15.2 (2016): [14] MIKOLOV, Tomas, et al. Efficient estimation of word representations in vector space. arxiv preprint arxiv: , [15] TensorFlow [16] Kingma, Diederik P., and Jimmy Ba. Adam: A method for stochastic optimization. arxiv preprint arxiv: (2014). [17] SZEGEDY, Christian, et al. Rethinking the inception architecture for computer vision. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition p [18] Levenshtein, Vladimir I. Binary codes capable of correcting deletions, insertions, and reversals. Soviet physics doklady. Vol. 10. No [7] ianchao Wu, Momo Klyen, Kazushige Ito, Zhan Chen Haiku Generation Using Deep Neural Networks 23 [8] Tosa, Naoko, Hideto Obara, and Michihiko Minoh. Hitch haiku: An interactive supporting system for composing haiku poem. International Conference on Entertainment Computing. Springer, Berlin, Heidelberg, [9] OPEN Hammerhead V1. [10] CC BY [11] [12] [13] Sundermeyer, Martin, Ralf Schlter, and Hermann Ney. LSTM neural networks for language modeling. Thirteenth Annual Conference of the International Speech Communication Association

IPSJ SIG Technical Report Vol.2017-CVIM-207 No /5/10 GAN 1,a) 2,b) Generative Adversarial Networks GAN GAN CIFAR-10 10% GAN GAN Stacked GAN Sta

IPSJ SIG Technical Report Vol.2017-CVIM-207 No /5/10 GAN 1,a) 2,b) Generative Adversarial Networks GAN GAN CIFAR-10 10% GAN GAN Stacked GAN Sta 1,a) 2,b) Generative Adversarial Networks CIFAR-10 10% Stacked Stacked 8.9% CNN 1. ILSVRC 1000 50000 5000 Convolutional Neural Network(CNN) [3] Stacked [4] 1 2 a) [email protected] b) [email protected]