Research map / Continual learning / Class-incremental
Generalized and Incremental Few-Shot Learning by Explicit Learning and Calibration Without Forgetting
Anna Kukleva, Hilde Kuehne, Bernt Schiele
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 |
|---|---|---|---|
| freeze batch norm during phases 2 and 3 (ablation) | LCwoF · mini-ImageNet 5w1s | hm_/J: 56.66 -> 57.83 | Improved |
| use cosine normalization of embedding and weights (ablation) | LCwoF · mini-ImageNet 5w1s | hm_/J: 57.83 -> 53.15 | Hurt |
| add a bias term to the linear classifier (ablation) | LCwoF · mini-ImageNet 5w1s | hm_/J: 57.83 -> 57.14 | Hurt |
| replace L2 weight constraints with knowledge distillation (ablation) | LCwoF · mini-ImageNet 5w1s and 5w5s | hm_/J: 57.83 -> 53.74 (5w1s); 66.61 -> 67.36 (5w5s) | Mixed |
| add unlabeled images to KD (KD+) (ablation) | LCwoF with KD · mini-ImageNet 5w1s | hm_/J: 53.74 -> 54.52 | Improved |
| add KD on top of L2 weight constraints (ablation) | LCwoF · mini-ImageNet 5w1s | hm_/J: 57.83 -> 56.75 | Hurt |
| lower the weight-constraint lambda to 1e+4 instead of the default (ablation) | LCwoF · mini-ImageNet 5w1s (phase 2 only) | hm_/J: 57.83 -> 50.78 | Hurt |
| add the full second phase instead of skipping it with batch ratio 5+2 (ablation) | LCwoF · mini-ImageNet 5w1s | hm_/J: 55.26 -> 57.83 | Improved |
| separate phases 2 and 3 instead of interleaving 10+1 (ablation) | LCwoF · mini-ImageNet 5w1s | hm_/J: 53.35 -> 57.83 | Improved |
| use unlimited base samples in phase 3 (ablation) | LCwoF · mini-ImageNet 5w1s ResNet | hm_/J: 57.39 -> 57.75 | Improved |
| use unlimited base samples in phase 3 | LCwoF · mini-Kinetics 5w1s | hm_/J: 49.12 -> 48.93 | Hurt |
| increase phase-2 duration from 30 to 150 epochs (ablation) | LCwoF · mini-ImageNet | accuracy | No clear effect |
Closest papers by meaning
- Few-Shot Class-Incremental Learning via Training-Free Prototype Calibration (NeurIPS 2023 · similarity 0.91)
- Learnable Distribution Calibration for Few-Shot Class-Incremental Learning (TPAMI 2023 · similarity 0.91)
- Overcoming Catastrophic Forgetting in Incremental Few-Shot Learning by Finding Flat Minima (NeurIPS 2021 · similarity 0.90)
- Rethinking Few-shot Class-incremental Learning: Learning from Yourself (ECCV 2024 · similarity 0.90)
- Few-Shot Class-Incremental Learning by Sampling Multi-Phase Tasks (TPAMI 2022 · similarity 0.90)
- Adaptive Decision Boundary for Few-Shot Class-Incremental Learning (AAAI 2025 · similarity 0.88)
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.
