Research map / Reinforcement learning / Visual and representation RL
Recurrent Hypernetworks are Surprisingly Strong in Meta-RL
Jacob Beck, Risto Vuorio, Zheng Xiong, Shimon Whiteson
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 |
|---|---|---|---|
| add a hypernetwork to the recurrent policy | RNN meta-RL · Cheetah-Dir | meta-episode return | Improved |
| add a hypernetwork to the recurrent policy | RNN meta-RL · Ant-Dir | meta-episode return | Improved |
| add a hypernetwork to the recurrent policy | RNN meta-RL · Cheetah-Vel | meta-episode return | Improved |
| add a hypernetwork to the recurrent policy | RNN meta-RL · Walker | meta-episode return | No clear effect |
| add a hypernetwork to the recurrent policy | RNN meta-RL · Grid-World, Grid-World Show, and Grid-World Dense | meta-episode return | Improved |
| add a hypernetwork to the recurrent policy | RNN meta-RL · MC-LS | meta-episode return | Improved |
| add a hypernetwork to the recurrent policy | RNN meta-RL · Meta-World ML10 | final average test success percent: 17.22 vs 13.87 | No clear effect |
| pass the current state directly to the policy, without a hypernetwork (ablation) | RNN · Grid-worlds and MuJoCo tasks | meta-episode return | Improved |
| use a hypernetwork in addition to directly passing the current state to the policy (ablation) | RNN+S · Grid-worlds and MuJoCo tasks | meta-episode return | Improved |
| replace Bias-HyperInit with Kaiming initialization (ablation) | RNN+HN · Walker | meta-episode return | Hurt |
| add a task-inference objective to end-to-end meta-RL (ablation) | RNN+HN and RNN+S · Grid-World | meta-episode return | Hurt |
| vary the relative weight of the task-inference objective (ablation) | combined end-to-end and task-inference methods · Grid-World | meta-episode return | No clear effect |
Closest papers by meaning
- Efficient Recurrent Off-Policy RL Requires a Context-Encoder-Specific Learning Rate (NeurIPS 2024 · similarity 0.88)
- Real-Time Recurrent Learning using Trace Units in Reinforcement Learning (NeurIPS 2024 · similarity 0.87)
- Exploring the Promise and Limits of Real-Time Recurrent Learning (ICLR 2024 · similarity 0.87)
- Reinforcement Learning Finetunes Small Subnetworks in Large Language Models (NeurIPS 2025 · similarity 0.87)
- 1000 Layer Networks for Self-Supervised RL: Scaling Depth Can Enable New Goal-Reaching Capabilities (NeurIPS 2025 · similarity 0.86)
- Reinforcement Learning for Reasoning in Large Language Models with One Training Example (NeurIPS 2025 · similarity 0.86)
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.
