the natural language of actions
play

The Natural Language of Actions Guy Tennenholtz , Shie Mannor ICML - PowerPoint PPT Presentation

The Natural Language of Actions Guy Tennenholtz , Shie Mannor ICML 2019 Technion Institute of Technology What is Language? What is Language? Language is a purely human and non-instinctive method of communicating ideas, emotions, and desires


  1. The Natural Language of Actions Guy Tennenholtz , Shie Mannor ICML 2019 Technion Institute of Technology

  2. What is Language?

  3. What is Language? ‘Language is a purely human and non-instinctive method of communicating ideas, emotions, and desires by means of voluntarily produced symbols.’ Edward Sapir, 1921 2

  4. What is Language? 3

  5. Contextual Semantics Contextual Representation A word’s contextual representation is an abstract congnitive structure that emmulates from encounters with the word in various linguistic contexts. We learn new words based on contextual cues ∙ I saw a little yazuba sleeping behind the tree. ∙ One glass of feandra is enough to get you drunk. 4

  6. Meaning of Actions? Characteristics ∙ Actions characterized by the company they keep ∙ Context is generated by an optimal agent ∙ Agent demonstrates acceptable behavior in the environment 5

  7. Act2Vec

  8. Act2Vec 7

  9. Skip Gram (used in Word2Vec) 8

  10. Skip Gram (used in Word2Vec) 9

  11. Act2Vec: Drawing ∙ 70,000 human drawn squares from the QuickDraw! dataset ∙ Strokes correspond to two operations 10

  12. Act2Vec: Navigation Act2Vec embedding trained in 2d environment on actions sequences of primitive actions: ∙ Move Forward 1 unit ∙ Turn Left 15 degrees ∙ Turn Right 15 degrees 11

  13. Act2Vec: Navigation 12

  14. Act2Vec in Reinforcement Learning

  15. Act2Vec in Reinforcement Learning 14

  16. Act2Vec for Q-Value Approximation 15

  17. Q-Embedding 1 1 He, Ji, et al. ”Deep Reinforcement Learning with a Natural Language Action Space.” Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (2016). 16

  18. Act2Vec for Exploration 17

  19. k-Exp k-Exp ∙ Divide action embedding space into k clusters using a clustering algorithm (e.g., k -means) ∙ Sample a cluster uniformly ∙ Given a cluster, uniformly sample an action within it 18

  20. Act2Vec for Domain Transfer 19

  21. Navigation Transfer to 3d environment 20

  22. The Semantics of Actions in Starcraft II

  23. StarCraft II 22

  24. StarCraft II 23

  25. StarCraft II 24

  26. Conclusion

  27. Conclusion ∙ Actions are thought of as symbols of a natural, expressive language ∙ Actions represent thoughts, beliefs, and strategies ∙ Their meaning is captured through their context ∙ Similarity between actions helps reduce cardinality of action space ∙ Prior knowledge is incorporated through action representations 26

Recommend


More recommend