Research map / Reinforcement learning / Visual and representation RL
Generalization of Reinforcement Learners with Working and Episodic Memory
Meire Fortunato, Melissa Tan, Ryan Faulkner, Steven Hansen, Adrià Puigdomènech Badia, Gavin Buttimore, Charles Deck, Joel Z. Leibo, Charles Blundell
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 episodic memory (MEM) + CPC to LSTM (ablation) | LSTM (IMPALA) · Memory Task Suite avg, Holdout-Extrapolate | human-normalized score: 35.6 -> 52.6 | Improved |
| add MEM to LSTM (ablation) | LSTM · Memory Task Suite avg, Holdout-Interpolate | human-normalized score: 40.2 -> 50.0 | Improved |
| add CPC to LSTM + MEM (ablation) | LSTM + MEM · Memory Task Suite avg, Train | human-normalized score: 78.7 -> 92.9 | Improved |
| add MEM to FF controller (ablation) | FF · Memory Task Suite avg, Train | human-normalized score: 33.9 -> 42.3 | Improved |
| add MEM + CPC to FF controller (ablation) | FF · Memory Task Suite avg, Train | human-normalized score: 33.9 -> 62.6 | Improved |
| replace CPC with reconstruction loss (REC) (ablation) | LSTM + MEM · Memory Task Suite avg, Train | human-normalized score: 92.9 -> 82.2 | Hurt |
| remove jumpy backpropagation from MRA (ablation) | MRA · Memory Suite tasks | task performance | Mixed |
| add MEM alone to FF controller (ablation) | FF · Change Detection Train | episode reward: 26.40 -> 25.76 | No clear effect |
| add MEM alone to FF controller (ablation) | FF · Continuous Recognition Train | episode reward: 26.90 -> 26.51 | No clear effect |
| add CPC to FF + MEM (ablation) | FF + MEM · Continuous Recognition Train | episode reward: 26.51 -> 49.60 | Improved |
| add CPC to FF + MEM (ablation) | FF + MEM · Change Detection Train | episode reward: 25.76 -> 44.76 | Improved |
| use REC instead of CPC with FF + MEM (ablation) | FF + MEM · Change Detection Train | episode reward: 44.76 -> 25.82 | Hurt |
Closest papers by meaning
- Benchmarking Agent Memory in Interdependent Multi-Session Agentic Tasks (ICML 2026 · similarity 0.88)
- Meta-Cognitive Memory Policy Optimization for Long-Horizon LLM Agents (NeurIPS 2026 · similarity 0.88)
- From Player to Master: Enhancing Test-Time Learning of LLM Agents via Reinforcement Learning over Memory (ICML 2026 · similarity 0.88)
- MemGen: Weaving Generative Latent Memory for Self-Evolving Agents (ICLR 2026 · similarity 0.87)
- Evaluating Memory in LLM Agents via Incremental Multi-Turn Interactions (ICLR 2026 · similarity 0.87)
- Investigating Pre-Training Objectives for Generalization in Vision-Based Reinforcement Learning (ICML 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.
