Research map / Continual learning / Online and replay-based
Few-shot Continual Infomax Learning
Ziqi Gu, Chunyan Xu, Jian Yang, Zhen Cui
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 feature embedding infomax (FEI) (ablation) | FCIL · miniImageNet (FSCIL) | Avg: 61.16 -> 62.13 | Improved |
| add continual structure infomax (CSI) (ablation) | FCIL · miniImageNet (FSCIL) | KR: 67.86 -> 68.61 | Improved |
| add both FEI and CSI (ablation) | FCIL · miniImageNet (FSCIL) | Avg: 61.16 -> 62.37 | Improved |
| add CSI in early sessions (t=2,3,4) (ablation) | FCIL · miniImageNet (FSCIL) | accuracy | No clear effect |
| add FCIL modules to FACT | FACT · miniImageNet (FSCIL) | final session accuracy: 49.48 -> 50.34 | Improved |
| add FCIL modules to CEC | CEC · miniImageNet (FSCIL) | final session accuracy: 47.63 -> 48.62 | Improved |
| swap KL for JSD in MI estimation, or cosine for KNN in relation construction (ablation) | FCIL · miniImageNet (FSCIL) | Avg: 62.37 -> 62.28 (JSD+Cos), 62.35 (KL+KNN) | No clear effect |
| use more convolutional feature levels L in feature embedding infomax (ablation) | FCIL · miniImageNet (FSCIL) | accuracy | No clear effect |
| increase number of shots in continual sessions (ablation) | FCIL · miniImageNet (FSCIL) | accuracy | Improved |
| increase number K of previous structures constrained in CSI (ablation) | FCIL · miniImageNet (FSCIL) | KR | Improved |
| FCIL versus FACT and CEC with matched base-session accuracy | FCIL · miniImageNet (FSCIL, equal base accuracy) | final session accuracy: 47.91 (FACT) -> 48.74 | Improved |
Closest papers by meaning
- Re-evaluating Continual Learning with Few-Shot Adaptation (NeurIPS 2026 · similarity 0.89)
- Continual Learning via Bit-Level Information Preserving (CVPR 2021 · similarity 0.88)
- Few-Shot and Continual Learning With Attentive Independent Mechanisms (ICCV 2021 · similarity 0.88)
- Stabilizing Zero-Shot Prediction: A Novel Antidote to Forgetting in Continual Vision-Language Tasks (NeurIPS 2024 · similarity 0.87)
- Do Your Best and Get Enough Rest for Continual Learning (CVPR 2025 · similarity 0.87)
- Pre-trained Vision and Language Transformers Are Few-Shot Incremental Learners (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.
