Research map / Reinforcement learning / Visual and representation RL
What Can Learned Intrinsic Rewards Capture?
Zeyu Zheng, Junhyuk Oh, Matteo Hessel, Zhongwen Xu, Manuel Kroiss, Hado Van Hasselt, David Silver, Satinder Singh
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use the learned intrinsic reward instead of extrinsic rewards or count-based exploration | new agents in Empty Rooms · Empty Rooms | episode return | Improved |
| use the learned intrinsic reward to explore uncertain object values and then exploit the better object | agents in Random ABC · Random ABC | episode return | Improved |
| use the learned intrinsic reward to discover the key-box relationship | agents in Key-Box · Key-Box | episode return | Improved |
| use the learned intrinsic reward to prepare for recurring task changes | agents in Non-stationary ABC · Non-stationary ABC | episode return | Improved |
| replace the lifetime-history reward input with current-timestep input (ablation) | learned intrinsic reward · Empty Rooms, Random ABC, Key-Box, and Non-stationary ABC | episode return | Hurt |
| replace the lifetime-return objective with the episodic-return objective (ablation) | learned intrinsic reward · Random ABC | episode return | Hurt |
| replace the lifetime-return objective with the episodic-return objective (ablation) | learned intrinsic reward · Non-stationary ABC | episode return | Hurt |
| replace the lifetime-return objective with the episodic-return objective (ablation) | learned intrinsic reward · Key-Box | episode return | Hurt |
| replace the lifetime-return objective with the episodic-return objective (ablation) | learned intrinsic reward · Empty Rooms | episode return | No clear effect |
| train an agent using the learned intrinsic reward rather than a MAML-transferred policy | agents in the paper's grid-world tasks · grid-world tasks | episode return | Mixed |
| apply the learned intrinsic reward to agents with permuted or extended actions | agents in Random ABC · Random ABC | episode return | Improved |
| apply the actor-critic-trained intrinsic reward to a Q-learning agent | Q-learning agent · Random ABC | episode return | Improved |
Closest papers by meaning
- On Learning Intrinsic Rewards for Policy Gradient Methods (NeurIPS 2018 · similarity 0.90)
- Instability of Meta-Learning Intrinsic Rewards for Policy Gradient Reinforcement Learning (NeurIPS 2026 · similarity 0.88)
- Discovering Temporally-Aware Reinforcement Learning Algorithms (ICLR 2024 · similarity 0.88)
- Successor-Predecessor Intrinsic Exploration (NeurIPS 2023 · similarity 0.87)
- Provably Efficient Learning of Transferable Rewards (ICML 2021 · similarity 0.87)
- Revisiting Intrinsic Reward for Exploration in Procedurally Generated Environments (ICLR 2023 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
