Research map / Label-efficient and robust learning / Domain adaptation and generalization
Less-Forgetful Learning for Domain Expansion in Deep Neural Networks
Heechul Jung, Jeongwoo Ju, Minju Jung, Junmo Kim
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 LF instead of modified LwF | Domain expansion learning · MNIST ∪ SVHN | Avg. (%): 89.28 -> 90.58 | Improved |
| Use LF instead of modified LwF | Domain expansion learning · CIFAR Color ∪ CIFAR Gray | Avg. (%): 74.33 -> 74.77 | Improved |
| Use LF instead of modified LwF | Domain expansion learning · ImageNet Normal ∪ Dark & Bright | Avg. (%): 83.56 -> 84.69 | Improved |
| Use LF instead of fine-tuning | Domain expansion learning · CIFAR-10 continual learning | Classification rate: 67.18 -> 71.1 | Improved |
| Use LF instead of offline training | CIFAR-10 continual learning · CIFAR-10 continual learning | Classification rate: 78.16 -> 71.1 | Hurt |
| Increase the feature-matching loss weight λe from 2.0×10^-4 to 1.6×10^-3 (ablation) | LF · MNIST ∪ SVHN | Avg. (%): 86.91 -> 90.58 | Improved |
| Increase the feature-matching loss weight λe from 2.0×10^-4 to 1.6×10^-3 (ablation) | LF · CIFAR Color ∪ CIFAR Gray | Avg. (%): 73.51 -> 74.77 | Improved |
| Vary λe across the tested values (ablation) | LF · ImageNet Normal ∪ Dark & Bright | Avg. (%): 84.04; 84.69; 84.34; 84.30 | Mixed |
| Replace ReLU with Maxout (ablation) | Fine-tuning · Tiny image experiments | Old/new domain classification rates | Mixed |
| Replace ReLU with LWTA (ablation) | Fine-tuning · Tiny image experiments | Old/new domain classification rates | No clear effect |
| Fine-tune only the softmax classifier layer (ablation) | Domain expansion learning · MNIST ∪ SVHN | New (%): 52.01 | Hurt |
| Use weight constraint instead of LF (ablation) | Domain expansion learning · MNIST ∪ SVHN | Old (%): 97.37 vs 80.29 | Hurt |
Closest papers by meaning
- Unsupervised Domain Adaptation with Residual Transfer Networks (NeurIPS 2016 · similarity 0.87)
- Selective Unlearning via Representation Erasure Using Domain Adversarial Training (ICLR 2025 · similarity 0.87)
- On Generalizing Beyond Domains in Cross-Domain Continual Learning (CVPR 2022 · similarity 0.86)
- Generalized Domain Conditioned Adaptation Network (TPAMI 2021 · similarity 0.86)
- Domain-Adaptive Deep Network Compression (ICCV 2017 · similarity 0.86)
- On Learning Invariant Representations for Domain Adaptation (ICML 2019 · 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.
