深層学習が示唆するend-to-end強化学習に基づく 機能創発アプローチの重要性と思考の創発に向けたカオスニューラルネットを用いた新しい強化学習

DOI Web Site

書誌事項

タイトル別名
  • Significance of Function Emergence Approach based on End-to-end Reinforcement Learning as suggested by Deep Learning, and Novel Reinforcement Learning Using a Chaotic Neural Network toward Emergence of Thinking
  • シンソウ ガクシュウ ガ シサ スル end-to-end キョウカ ガクシュウ ニ モトズク キノウソウハツ アプローチ ノ ジュウヨウセイ ト シコウ ノ ソウハツ ニ ムケタ カオスニューラルネット オ モチイタ アタラシイ キョウカ ガクシュウ

この論文をさがす

説明

 It is propounded that in order to avoid the “frame problem” or “symbol grounding<br> problem” and to create a way to analyze and realize human-like intelligence with higher<br> functions, it is not enough just to introduce deep learning, but it is significant to get<br> out of deeply penetrated “division into functional modules” and to take the approach of<br> “function emergence through end-to-end reinforcement learning.” The functions that<br> have been shown to emerge according to this approach in past works are summarized,<br>and the reason for the difficulty in the emergence of thinking that is a typical higher<br> function is made clear.<br>  It is claimed that the proposed hypothesis that exploration grows towards think-<br>ing through learning, becomes a key to break through the difficulty. To realize that,<br>“reinforcement learning using a chaotic neural network” in which adding external ex-<br>ploration noises is not necessary is introduced. It is shown that a robot with two<br> wheels and a simple visual sensor can learn an obstacle avoidance task by using this<br> new reinforcement learning method.

収録刊行物

  • 認知科学

    認知科学 24 (1), 96-117, 2017

    日本認知科学会

関連プロジェクト

もっと見る

キーワード

詳細情報 詳細情報について

問題の指摘

ページトップへ