Research map / Reinforcement learning / Visual and representation RL
Transfer in Deep Reinforcement Learning Using Successor Features and Generalised Policy Improvement
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 SF&GPI continual learning instead of SF&GPI transfer alone | SF&GPI on DeepMind Lab 3D object room test tasks · DeepMind Lab test tasks | average reward per episode | Helped |
| Use SF&GPI transfer instead of learning a task-specific policy from scratch | DeepMind Lab 3D object room test tasks · Unseen test tasks | average reward per episode | Mixed |
| Use a linearly dependent base-task set that does not span the test rewards (ablation) | SF&GPI-transfer · DeepMind Lab test tasks | average reward per episode | Mixed |
| Use non-canonical base tasks that span the task space (ablation) | SF&GPI-transfer · Nine DeepMind Lab test tasks | average reward per episode | Mixed |
| Add a specialized policy for the current task and include it in GPI (ablation) | SF&GPI-transfer · DeepMind Lab test tasks with negative rewards | average reward per episode | Helped |
| Add a specialized policy for the current task and include it in GPI (ablation) | SF&GPI-transfer · DeepMind Lab test tasks with only positive rewards | average reward per episode | No clear effect |
Closest papers by meaning
- Successor Features for Transfer in Reinforcement Learning (NeurIPS 2017 · similarity 0.92)
- SF-DQN: Provable Knowledge Transfer using Successor Feature for Deep Reinforcement Learning (ICML 2024 · similarity 0.91)
- Learning Successor Features the Simple Way (NeurIPS 2024 · similarity 0.89)
- Contextual Pre-planning on Reward Machine Abstractions for Enhanced Transfer in Deep Reinforcement Learning (AAAI 2024 · similarity 0.88)
- Distributional Successor Features Enable Zero-Shot Policy Optimization (NeurIPS 2024 · similarity 0.87)
- Model-Based Transfer Learning for Contextual Reinforcement Learning (NeurIPS 2024 · 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.