Machine vs Human Attention in Deep Reinforcement Learning Tasks
Dec 6, 2021
Speakers
Sihang Guo
Speaker · 0 followers
Ruohan Zhang
Speaker · 0 followers
Yifeng Zhu
Speaker · 0 followers
Dana Ballard
Speaker · 0 followers
Peter Stone
Speaker · 4 followers
About
Deep reinforcement learning (RL) algorithms are powerful tools for solving visuomotor decision tasks. However, the trained models are often difficult to interpret, because they are represented as end-to-end deep neural networks. In this paper, we shed light on the inner workings of such trained models by analyzing the pixels that they attend to during task execution, and comparing them with the pixels attended to by humans executing the same tasks. To this end, we investigate the following two questions that, to the best of our knowledge, have not been previously studied. 1) How similar are the visual features learned by RL agents and humans when performing the same task? and, 2) How do similarities and differences in these learned features explain RL agents' performance on these tasks? Specifically, we compare the saliency maps of RL agents against visual attention models of human experts when learning to play Atari games. Further, we analyze how hyperparameters of the deep RL algorithm affect the learned features and saliency maps of the trained agents. The insights provided have the potential to inform novel algorithms for closing the performance gap between human experts and RL agents.Deep reinforcement learning (RL) algorithms are powerful tools for solving visuomotor decision tasks. However, the trained models are often difficult to interpret, because they are represented as end-to-end deep neural networks. In this paper, we shed light on the inner workings of such trained models by analyzing the pixels that they attend to during task execution, and comparing them with the pixels attended to by humans executing the same tasks. To this end, we investigate the following two q…
Organizer
NeurIPS 2021
Account · 1.9k followers
About NeurIPS 2021
Neural Information Processing Systems (NeurIPS) is a multi-track machine learning and computational neuroscience conference that includes invited talks, demonstrations, symposia and oral and poster presentations of refereed papers. Following the conference, there are workshops which provide a less formal setting.
Like the format? Trust SlidesLive to capture your next event!
Professional recording and live streaming, delivered globally.
Sharing
Recommended Videos
Presentations on similar topic, category or speaker
The Neurobiology of Pragmatics
Watch later
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%
MHER: Model-based Hindsight Experience Replay
Watch later
Ray Rui, …
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%
Federated Linear Contextual Bandits
Watch later
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%
Gradient-matching coresets for continual learning
Watch later
Lukas Balles, …
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%
On the Theory of Reinforcement Learning with Once-per-Episode Feedback
Watch later
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%
Sculpting (human-like) AI systems by sculpting their (social) environments
Watch later
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%