論文紹介:Dueling network architectures for deep reinforcement learningKazuki Adachi
?
Wang, Ziyu, et al. "Dueling network architectures for deep reinforcement learning." Proceedings of The 33rd International Conference on Machine Learning, PMLR 48:1995-2003, 2016.
ERATO感謝祭 Season IV
【参考】Satoshi Hara and Takanori Maehara. Enumerate Lasso Solutions for Feature Selection. In Proceedings of the 31st AAAI Conference on Artificial Intelligence (AAAI'17), pages 1985--1991, 2017.
Deep Learningについて、日本情報システム?ユーザー協会(JUAS)のJUAS ビジネスデータ研究会 AI分科会で発表しました。その際に使用した資料です。専門家向けではなく、一般向けの資料です。
なお本資料は、2015年12月の日本情報システム?ユーザー協会(JUAS)での発表資料の改訂版となります。
This document discusses generative adversarial networks (GANs) and their relationship to reinforcement learning. It begins with an introduction to GANs, explaining how they can generate images without explicitly defining a probability distribution by using an adversarial training process. The second half discusses how GANs are related to actor-critic models and inverse reinforcement learning in reinforcement learning. It explains how GANs can be viewed as training a generator to fool a discriminator, similar to how policies are trained in reinforcement learning.
(DL Hacks輪読) How transferable are features in deep neural networks?Masahiro Suzuki
?
This document summarizes an experiment on measuring how transferable features are in deep neural networks. The experiment trained neural networks on halves of the ImageNet dataset and tested how well the networks could generalize to the other half. It found that earlier layer features transferred better than later layer features, and that fine-tuning improved performance. Transferring between more dissimilar datasets led to poorer performance. Randomly initialized weights performed worse than trained weights.
ERATO感謝祭 Season IV
【参考】Satoshi Hara and Takanori Maehara. Enumerate Lasso Solutions for Feature Selection. In Proceedings of the 31st AAAI Conference on Artificial Intelligence (AAAI'17), pages 1985--1991, 2017.
Deep Learningについて、日本情報システム?ユーザー協会(JUAS)のJUAS ビジネスデータ研究会 AI分科会で発表しました。その際に使用した資料です。専門家向けではなく、一般向けの資料です。
なお本資料は、2015年12月の日本情報システム?ユーザー協会(JUAS)での発表資料の改訂版となります。
This document discusses generative adversarial networks (GANs) and their relationship to reinforcement learning. It begins with an introduction to GANs, explaining how they can generate images without explicitly defining a probability distribution by using an adversarial training process. The second half discusses how GANs are related to actor-critic models and inverse reinforcement learning in reinforcement learning. It explains how GANs can be viewed as training a generator to fool a discriminator, similar to how policies are trained in reinforcement learning.
(DL Hacks輪読) How transferable are features in deep neural networks?Masahiro Suzuki
?
This document summarizes an experiment on measuring how transferable features are in deep neural networks. The experiment trained neural networks on halves of the ImageNet dataset and tested how well the networks could generalize to the other half. It found that earlier layer features transferred better than later layer features, and that fine-tuning improved performance. Transferring between more dissimilar datasets led to poorer performance. Randomly initialized weights performed worse than trained weights.
31. Computer Aided Design t = [0,1,2,3,4,5]
B-Spline ノットは1ずつ単純増加
t - ti
ti+1 - ti bi,0
t - ti
bi,1
ti+2 - ti ti+2 - t bi+1,0
ti+2 - ti+1
bi,2
t - ti+1
ti+2 - ti+1 bi+1,0
ti+3 - t
bi+1,1
ti+3 - ti+1 ti+3 - t
bi+2,0
ti+3 - ti+2
32. Computer Aided Design
B-Spline
t - ti bi,0
t - ti
bi,1
2
ti+2 - t bi+1,0
bi,2
t - ti+1 bi+1,0
ti+3 - t
bi+1,1
2
ti+3 - t bi+2,0
33. Computer Aided Design
B-Spline
i=0のとき
t b0,0
t
bi,1
2
2-t b1,0
b0,2
t - 1 b1,0
3-t
bi+1,1
2
3 - t b2,0
34. Computer Aided Design
B-Spline
i=0 0 t<1のとき t2
t b0,0
t 2
bi,1
2
2-t b1,0
b0,2
t - 1 b1,0
3-t
bi+1,1
2
3 - t b2,0
35. Computer Aided Design
B-Spline
i=0 1 t<2のとき
t b0,0
t
bi,1
2
2-t b1,0
2
3
b0,2 -t + 3t - 2
t - 1 b1,0
3-t
bi+1,1
2
3 - t b2,0
36. Computer Aided Design
B-Spline
i=0 2 t<3のとき
t b0,0
t
bi,1
2
2-t b1,0
b0,2
t - 1 b1,0
3-t
bi+1,1 t 2 - 6t + 9
2
3 - t b2,0 2
46. t - ti
ti+1 - ti bi,0
Computer Aided Design t - ti
bi,1
ti+2 - ti ti+2 - t bi+1,0
Cubic B-Spline ti+2 - ti+1
t - ti
bi,2
ti+3 - t
t - ti+1
ti+2 - ti+1
bi+1,0
ti+3 - t
bi+1,1
ti+3 - ti+1 ti+3 - t
ti+3 - ti+2
bi+2,0
bi,3
t - ti
ti+2 - ti bi+1,0
t - ti+1
bi+1,1
ti+3 - ti+1 ti+3 - t bi+2,0
ti+3 - ti+2
ti+4 - t bi+1,2
ti+4 - ti+1
t - ti+2
ti+3 - ti+2
bi+2,0
ti+4 - t
bi+2,1
ti+4 - ti+2 ti+4 - t
ti+4 - ti+3
bi+3,0