Research map / Reinforcement learning / Visual and representation RL
Distral: Robust multitask reinforcement learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 a shared distilled policy to regularize task-specific policies | Multitask reinforcement learning · Two-room grid world | learning speed | Helped |
| Use Distral instead of multitask A3C | DeepMind Lab multitask RL · Mazes | final performance | Helped |
| Use a two-column policy architecture instead of the corresponding single-column architecture (ablation) | Distral · Mazes | learning speed | Helped |
| Remove the additional entropy regularization (ablation) | Distral · Mazes | final score | Hurt |
| Remove KL regularization (ablation) | Distral · DeepMind Lab tasks | learning stability | Hurt |
| Use KL regularization without entropy regularization (ablation) | Distral distilled policy · Mazes | distilled-policy performance | Helped |
| Use a two-column architecture rather than a single-column architecture (ablation) | Distral · Navigation | final performance | Helped |
| Use Distral instead of multitask A3C | Multitask reinforcement learning · Laser-tag | final performance | Helped |
| Use a single-column rather than a two-column architecture (ablation) | Distral · Laser-tag | early performance | Helped |
| Add entropy regularization (ablation) | Distral · Laser-tag | learning progress | Helped |
| Use separate task-specific policies instead of a shared multitask policy (ablation) | A3C · Laser-tag | final performance | Helped |
| Use the original A3C method rather than entropy-regularized A3C (ablation) | A3C · DeepMind Lab tasks | performance | No clear effect |
Closest papers by meaning
- DaCe-DT: Data-Centric Offline Multi-Task Reinforcement Learning via Adaptive Prompts and Trajectory Correction for Heterogeneous Tasks (NeurIPS 2026 · similarity 0.86)
- Task Adaptation from Skills: Information Geometry, Disentanglement, and New Objectives for Unsupervised Reinforcement Learning (ICLR 2024 · similarity 0.85)
- Robust Policy Learning via Offline Skill Diffusion (AAAI 2024 · similarity 0.85)
- Multi-Task Reinforcement Learning with Mixture of Orthogonal Experts (ICLR 2024 · similarity 0.85)
- Offline Multitask Representation Learning for Reinforcement Learning (NeurIPS 2024 · similarity 0.85)
- QMP: Q-switch Mixture of Policies for Multi-Task Behavior Sharing (ICLR 2025 · similarity 0.85)
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.