Research map / Continual learning / Class-incremental
Efficient Feature Transformations for Discriminative and Generative Continual Learning
Vinay Kumar Verma, Kevin J Liang, Nikhil Mehta, Piyush Rai, Lawrence Carin
Check 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 |
|---|---|---|---|
| replace prior continual-learning methods with EFT feature transforms | ResNet-18 · CIFAR-100/10 CIL | final average accuracy: 36.2 -> 45.5 | Improved |
| replace PackNet with EFT feature transforms | VGG-16 · Tiny-ImageNet-200/10 task incremental | average top-1 accuracy: 56.4 -> 66.8 | Improved |
| replace PackNet with EFT feature transforms | AlexNet · ImageNet-1K/10 task incremental | average top-1 accuracy: 59.1 -> 59.4 | Improved |
| add feature distance regularization (ablation) | EFT-a8b16 · CIFAR-100/10 | CIL accuracy: 44.4 -> 45.5 | Improved |
| add feature distance regularization (ablation) | EFT-a8b16 · CIFAR-100/10 | task prediction accuracy: 44.9 -> 46.1 | Improved |
| increase transform group sizes from a4b0 to a8b16 (ablation) | EFT · CIFAR-100/10 | accuracy: 44.4 -> 45.5 | Improved |
| use spatial transforms alongside pointwise transforms rather than pointwise transforms alone (ablation) | EFT · CIFAR-100/10 | accuracy: 43.2 -> 45.5 | Improved |
| initialize each new task's transform from the previous task's transform instead of randomly (ablation) | EFT-a8b16 · CIFAR-100/10 | accuracy | Improved |
| use parallel rather than sequential feature adaptation (ablation) | EFT · continual learning classification | accuracy | Hurt |
| use EFT instead of MeRGAN-RA | StackGAN-v2 · continual generation: cats, birds, churches, towers | final average FID: 42.8 -> 33.1 | Improved |
| replace APD with EFT | VGG-16 · heterogeneous sequence SVHN → CIFAR-10 → CIFAR-100 | average accuracy: 83.0 -> 83.5 | Improved |
| replace APD with EFT | VGG-16 · heterogeneous sequence CIFAR-100 → CIFAR-10 → SVHN | average accuracy: 85.0 -> 85.8 | Improved |
Closest papers by meaning
- A continual learning survey: Defying forgetting in classification tasks (TPAMI 2021 · similarity 0.89)
- The Importance of Being Lazy: Scaling Limits of Continual Learning (ICML 2025 · similarity 0.87)
- LoRanPAC: Low-rank Random Features and Pre-trained Models for Bridging Theory and Practice in Continual Learning (ICLR 2025 · similarity 0.87)
- LoRA Subtraction for Drift-Resistant Space in Exemplar-Free Continual Learning (CVPR 2025 · similarity 0.87)
- Optimal Task Order for Continual Learning of Multiple Tasks (ICML 2025 · similarity 0.87)
- Forgetting, Ignorance or Myopia: Revisiting Key Challenges in Online Continual Learning (NeurIPS 2024 · similarity 0.87)
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.
