Research map / Learning theory and algorithms / Prediction and decision losses
Efficient Lifelong Learning with A-GEM
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 |
|---|---|---|---|
| replace GEM's per-task constraints with one averaged episodic-memory constraint | GEM · Permuted MNIST | LCA_10: 0.23 (± 0.005) -> 0.29 (± 0.004) | Helped |
| replace GEM's per-task constraints with one averaged episodic-memory constraint | GEM · Split CIFAR | A_T (%): 61.2 (± 0.78) -> 62.3 (± 1.24) | Helped |
| replace GEM's per-task constraints with one averaged episodic-memory constraint | GEM · Split CIFAR | Training time [s]: 5238 -> 449 | Helped |
| replace GEM's per-task constraints with one averaged episodic-memory constraint | GEM · Split CIFAR | LCA_10: 0.36 (± 0.007) -> 0.35 (± 0.01) | Hurt |
| provide compositional task descriptors through a joint embedding model (ablation) | lifelong learning methods · Split CUB and Split AWA | few-shot performance | Helped |
| sample one GEM constraint randomly at each training step (ablation) | GEM · Permuted MNIST | A_T (%): 89.5 -> 88.2 | Hurt |
| sample one GEM constraint randomly at each training step (ablation) | GEM · Split CIFAR | A_T (%): 61.2 -> 56.2 | Hurt |
| use Progressive Networks instead of A-GEM | A-GEM · Permuted MNIST | A_T (%): 89.1 (± 0.14) -> 93.5 (± 0.07) | Helped |
| use joint embedding task descriptors (ablation) | A-GEM · Split CUB and Split AWA | A_T (%) | Helped |
Closest papers by meaning
- Improved Schemes for Episodic Memory-based Lifelong Learning (NeurIPS 2020 · similarity 0.87)
- Adaptive Retention & Correction: Test-Time Training for Continual Learning (ICLR 2025 · similarity 0.84)
- Active Lifelong Learning With "Watchdog" (AAAI 2018 · similarity 0.83)
- Lifelong Learning with Weighted Majority Votes (NeurIPS 2016 · similarity 0.83)
- A Statistical Theory of Regularization-Based Continual Learning (ICML 2024 · similarity 0.82)
- Efficient Lifelong Model Evaluation in an Era of Rapid Progress (NeurIPS 2024 · similarity 0.81)
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.