Research map / Continual learning / Online and replay-based
Generative vs. Discriminative: Rethinking The Meta-Continual Learning
Mohammad Amin Banayeean Zade, Rasoul Mirzaiezadeh, Hosein Hasani, Mahdieh Soleymani Baghshah
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 |
|---|---|---|---|
| use generative Bayesian classifier head instead of discriminative logistic regression (continual) (ablation) | MTLR · Mini-ImageNet 20-way 100-shot continual | accuracy: 31.1±1.1 -> 59.2±0.9 | Improved |
| use generative Bayesian classifier head instead of discriminative logistic regression (continual) (ablation) | MTLR · Omniglot 600-way 15-shot continual | accuracy: 61.9±0.7 -> 86.9±0.6 | Improved |
| update only class-specific weights (pseudo-generative LR) instead of all head weights (ablation) | MTLR · Mini-ImageNet 20-way 100-shot continual | accuracy: 31.1±1.1 -> 53.2±1.1 | Improved |
| move from IID to continual training of the classifier (ablation) | MTLR · Mini-ImageNet 20-way 100-shot | accuracy: 48.9±1.1 -> 31.1±1.1 | Hurt |
| move from IID to continual training of the classifier (ablation) | GeMCL · Mini-ImageNet 20-way 100-shot | accuracy: 59.2±0.9 -> 59.2±0.9 | No clear effect |
| replace pretrained feature extractor with meta-trained one (ablation) | GeMCL · Omniglot 600-way 15-shot continual | accuracy: 75.9±0.8 -> 86.9±0.6 | Improved |
| replace pretrained feature extractor with meta-trained one (ablation) | GeMCL · Mini-ImageNet 20-way 100-shot continual | accuracy: 55.9±1.1 -> 59.2±0.9 | Improved |
| replace pretrained feature extractor with meta-trained one (ablation) | MTLR · Mini-ImageNet 20-way 100-shot IID | accuracy: 60.9±1.0 -> 48.9±1.1 | Hurt |
| add Bayesian prior statistics and learned variance to prototypical classifier (ablation) | Prototypical · Omniglot and Mini-ImageNet (curves) | accuracy | Improved |
Closest papers by meaning
- Generative Continual Concept Learning (AAAI 2020 · similarity 0.91)
- Recall-Oriented Continual Learning with Generative Adversarial Meta-Model (AAAI 2024 · similarity 0.91)
- Learning to Continually Learn with the Bayesian Principle (ICML 2024 · similarity 0.90)
- Continual Learning with Deep Generative Replay (NeurIPS 2017 · similarity 0.90)
- Overcoming Catastrophic Forgetting by Bayesian Generative Regularization (ICML 2021 · similarity 0.89)
- A continual learning survey: Defying forgetting in classification tasks (TPAMI 2021 · similarity 0.89)
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.
