Research map / Continual learning / Online and replay-based
Modular Continual Learning in a Unified Visual Environment
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 the EMS architecture with an early bottleneck, multiplicative interactions, and symmetry instead of standard module motifs | ReMaP modules · 12 visual task variants | TA-N-AUC: 0.997 | Helped |
| Add low-order polynomial multiplicative interactions (ablation) | EMS ReMaP module · 12 visual task variants | TA-N-AUC: 0.818 | Helped |
| Add symmetric structure (ablation) | EMS ReMaP module · 12 visual task variants | TA-N-AUC: 0.944 | Helped |
| Add an early visual bottleneck, symmetry, and multiplicative interactions rather than using a large ReLU MLP (ablation) | ReMaP module · 12 visual task variants | TA-N-AUC: 0.717 | Helped |
| Add symmetry and multiplicative interactions (ablation) | ReMaP module · 12 visual task variants | TA-N-AUC: 0.566 | Helped |
| Add the EMS structure instead of using a parameter-matched, fully ablated small module (ablation) | ReMaP module · 12 visual task variants | TA-N-AUC: 0.403 | Helped |
| Use convolutional visual features with spatial information rather than FC6-only features (ablation) | EMS localization module · Localization and MS-COCO MTS | IoU | Helped |
| Use single-unit voting rather than layer voting | Dynamic Neural Voting controller · Task switching | Relative Gain in AUC | Helped |
| Reuse previously learned module pathways through dynamic neural voting rather than train the new task from scratch | Dynamic Neural Voting controller · 15 task switches | Relative Gain in AUC | Helped |
| Use EMS rather than a large fully ablated module with dynamic voting | Dynamic Neural Voting controller · Task switching | Transfer Gain | Helped |
| Allocate a new module and let the controller reuse an existing module when the task has not actually changed | Dynamic Neural Voting controller · No-switch task transitions | Reward | No clear effect |
Closest papers by meaning
- Efficient Continual Learning with Modular Networks and Task-Driven Priors (ICLR 2021 · similarity 0.90)
- Self-Composing Policies for Scalable Continual Reinforcement Learning (ICML 2024 · similarity 0.89)
- Principled Fast and Meta Knowledge Learners for Continual Reinforcement Learning (ICLR 2026 · similarity 0.88)
- A Probabilistic Framework for Modular Continual Learning (ICLR 2024 · similarity 0.87)
- MoRe: Modular Representations for Principled Continual Representation Learning on Sequential Data (NeurIPS 2026 · similarity 0.86)
- Optimal Task Order for Continual Learning of Multiple Tasks (ICML 2025 · similarity 0.86)
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.