Research map / Continual learning / Class-incremental
Split-and-Bridge: Adaptable Class Incremental Learning within a Single Neural Network
Jong-Yeong Kim, Dong-Wan Choi
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 Split-and-Bridge instead of the strongest listed baseline | ResNet-18 · CIFAR-100, 5 tasks | average accuracy: 67.68 -> 68.62 | Improved |
| Use Split-and-Bridge instead of the strongest listed baseline | ResNet-18 · CIFAR-100, 10 tasks | average accuracy: 63.72 -> 66.97 | Improved |
| Use Split-and-Bridge instead of the strongest listed baseline | ResNet-18 · CIFAR-100, 20 tasks | average accuracy: 58.08 -> 61.12 | Improved |
| Use Split-and-Bridge instead of the strongest listed baseline | ResNet-18 · Tiny-ImageNet, 2 tasks | average accuracy: 58.33 -> 60.52 | Improved |
| Use Split-and-Bridge instead of the strongest listed baseline | ResNet-18 · Tiny-ImageNet, 5 tasks | average accuracy: 56.80 -> 57.16 | Improved |
| Use Split-and-Bridge instead of the strongest listed baseline | ResNet-18 · Tiny-ImageNet, 10 tasks | average accuracy: 53.12 -> 54.81 | Improved |
| Use Split-and-Bridge instead of the strongest listed baseline | ResNet-18 · Tiny-ImageNet, 20 tasks | average accuracy: 48.14 -> 51.63 | Improved |
| Add KD loss to CE loss (ablation) | ResNet-18 · CIFAR-100, 2 tasks | overall accuracy: 74.25 -> 75.67 | Improved |
| Add KD loss to CE loss (ablation) | ResNet-18 · CIFAR-100, 2 tasks | new accuracy: 89.6 -> 85.75 | Hurt |
| Add KD loss to CE loss (ablation) | ResNet-18 · CIFAR-100, 2 tasks | intra-new accuracy: 90.85 -> 88.30 | Hurt |
| Separate old-task and new-task upper-layer branches during split-phase training | Split-and-Bridge · CIFAR-100 and Tiny-ImageNet | intra-new accuracy | Improved |
| Use two separately trained networks and combine them with double distillation | DD with WA · CIFAR-100 and Tiny-ImageNet | average accuracy | Mixed |
Closest papers by meaning
- Train with Perturbation, Infer after Merging: A Two-Stage Framework for Continual Learning (NeurIPS 2025 · similarity 0.86)
- Convergence and Implicit Bias of Gradient Descent on Continual Linear Classification (ICLR 2025 · similarity 0.86)
- Rethinking Momentum Knowledge Distillation in Online Continual Learning (ICML 2024 · similarity 0.86)
- MIND: Multi-Task Incremental Network Distillation (AAAI 2024 · similarity 0.86)
- Progressively Knowledge Distillation via Re-parameterizing Diffusion Reverse Process (AAAI 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.
