Research map / Continual learning / Online and replay-based
Lifelong Learning of Compositional Structures
Jorge A Mendez, ERIC EATON
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 |
|---|---|---|---|
| separate task-structure learning from component adaptation | lifelong compositional learning with soft layer ordering · MNIST | accuracy: 94.2±0.3% -> 96.5±0.2% | Improved |
| add dynamically expandable components (ablation) | ER Compositional soft layer ordering · CIFAR | accuracy: 58.7±0.5% -> 77.6±0.3% | Improved |
| add dynamically expandable components (ablation) | ER Compositional soft layer ordering · CUB | accuracy: 80.6±0.3% -> 79.0±0.5% | Mixed |
| add dynamically expandable components (ablation) | ER Compositional soft layer ordering · MNIST | accuracy: 96.5±0.2% -> 97.6±0.2% | Improved |
| adapt components after task assimilation rather than keeping all components fixed (ablation) | FM compositional learner · CIFAR | accuracy: 48.8±1.6% -> 58.7±0.5% | Improved |
| use compositional structures instead of a single shared monolithic architecture | ER lifelong learner · Combined all data sets | accuracy: 47.4±4.5% -> 86.5±1.8% | Improved |
| use one shared linear model rather than compositional linear models | ER linear learner · Schools | RMSE: 10.65±0.04 -> 10.34±0.02 | Improved |
| use soft gating rather than soft layer ordering | ER dynamic compositional learner · MNIST | accuracy: 97.6±0.2% -> 98.2±0.1% | Improved |
Closest papers by meaning
- Separating the 'what' and 'how' of compositional computation to enable reuse and continual learning (NeurIPS 2025 · similarity 0.88)
- Compositional Reasoning in Language Models under Reinforcement Learning Post-Training (NeurIPS 2026 · similarity 0.87)
- CraftFactory: A Conditioned Control Policy Benchmark for Compositional Generalization (AAAI 2025 · similarity 0.85)
- Compositional Generalization from Learned Skills via CoT Training: A Theoretical and Structural Analysis for Reasoning (ICLR 2026 · similarity 0.85)
- Discovering modular solutions that generalize compositionally (ICLR 2024 · similarity 0.85)
- Limits of Transformer Language Models on Learning to Compose Algorithms (NeurIPS 2024 · 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.
