Research map / Continual learning / Online and replay-based
Few-Shot and Continual Learning With Attentive Independent Mechanisms
Eugene Lee, Cheng-Han Huang, Chen-Yi Lee
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 AIM module to SIB | SIB (Conv-4-64) · MiniImageNet 5-way 1-shot | accuracy: 58.0 -> 61.90 | Improved |
| add AIM module to SIB | SIB (WRN-28-10) · MiniImageNet 5-way 1-shot | accuracy: 70.0 -> 71.22 | Improved |
| add AIM module to SIB | SIB (WRN-28-10) · CIFAR-FS 5-way 5-shot | accuracy: 85.3 -> 87.34 | Improved |
| replace AIM with a parameter-matched linear layer in SIB (ablation) | SIB + Linear (Conv-4-64) · MiniImageNet 5-way 1-shot | accuracy: 60.07 -> 61.90 | Improved |
| replace AIM with a parameter-matched linear layer in SIB (ablation) | SIB + Linear (WRN-28-10) · CIFAR-FS 5-way 5-shot | accuracy: 79.91 -> 87.34 | Improved |
| use soft (attention above 0.5) instead of hard top-K mechanism selection (ablation) | AIM (Conv-4-64) · MiniImageNet 5-way 1-shot | accuracy: 61.90 -> 61.74 | No clear effect |
| use soft (attention above 0.5) instead of hard top-K mechanism selection (ablation) | AIM (WRN-28-10) · MiniImageNet 5-way 1-shot | accuracy: 71.03 -> 69.96 | Hurt |
| use soft instead of hard mechanism selection (ablation) | AIM (WRN-28-10) · CIFAR-FS 5-way 1-shot | accuracy: 79.19 -> 80.14 | Improved |
| increase stochastic sampling count K+l from 10 to 32 (ablation) | AIM (Conv-4-64) · CIFAR-FS 5-way 1-shot | accuracy: 71.09 -> 66.90 | Hurt |
| increase active mechanisms K from 1 to 2 (ablation) | AIM (Conv-4-64) · MiniImageNet 5-way 1-shot | accuracy: 25.54 -> 62.00 | Improved |
| increase active mechanisms K from 2 to 32 (ablation) | AIM (Conv-4-64) · MiniImageNet 5-way 5-shot | accuracy: 74.62 -> 74.79 | No clear effect |
| add AIM to OML | OML · Omniglot (100 classes) | accuracy: 73.61 -> 97.75 | Improved |
Closest papers by meaning
- Few-shot Continual Infomax Learning (ICCV 2023 · similarity 0.88)
- Learning to Forget for Meta-Learning via Task-and-Layer-Wise Attenuation (TPAMI 2021 · similarity 0.88)
- Re-evaluating Continual Learning with Few-Shot Adaptation (NeurIPS 2026 · similarity 0.87)
- Stabilizing Zero-Shot Prediction: A Novel Antidote to Forgetting in Continual Vision-Language Tasks (NeurIPS 2024 · similarity 0.87)
- Pre-trained Vision and Language Transformers Are Few-Shot Incremental Learners (CVPR 2024 · similarity 0.86)
- Enhancing Visual Continual Learning with Language-Guided Supervision (CVPR 2024 · 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.
