Research map / Label-efficient and robust learning / Recognition and long-tail
Composing Task Knowledge With Modular Successor Feature Approximators
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 |
|---|---|---|---|
| Modular cumulants and SFs (MSFA) instead of entangled functions (ablation) | MSFA · BabyAI object navigation | generalization episode return | Helped |
| Removing GPI from SF-based methods (ablation) | USFA · BabyAI object navigation | generalization episode return | Hurt |
| Modular architecture without GPI (greedy SF policy) (ablation) | MSFA · BabyAI object navigation | generalization episode return | Helped |
| Learning cumulants instead of hand-designed (USFA-Learned-phi) (ablation) | USFA · BabyAI object navigation | generalization episode return | Hurt |
| Modular learned cumulants and SFs (MSFA) vs. USFA with hand-designed cumulants | MSFA · BabyAI object navigation | generalization episode return | No clear effect |
| Modular learned cumulants and SFs (MSFA) vs. learned-cumulant USFA | MSFA · Fruitbot (Procgen) collecting multiple objects | generalization return | Helped |
| Modular learned cumulants and SFs (MSFA) | MSFA · Fruitbot tasks requiring avoiding objects | generalization return | No clear effect |
| Modular SF architecture vs. UVFA-based methods | MSFA · Minihack heterogeneous tasks (large room) | success rate | Helped |
| SF-based methods vs UVFA in small Minihack room | MSFA · Minihack small room | success rate | No clear effect |
Closest papers by meaning
- Combining Behaviors with the Successor Features Keyboard (NeurIPS 2023 · similarity 0.89)
- Successor Feature Sets: Generalizing Successor Representations Across Policies (AAAI 2021 · similarity 0.88)
- SF-DQN: Provable Knowledge Transfer using Successor Feature for Deep Reinforcement Learning (ICML 2024 · similarity 0.85)
- Learning Successor Features the Simple Way (NeurIPS 2024 · similarity 0.84)
- Temporal Representation Alignment: Successor Features Enable Emergent Compositionality in Robot Instruction Following (NeurIPS 2025 · similarity 0.84)
- $\pi$2vec: Policy Representation with Successor Features (ICLR 2024 · similarity 0.83)
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.