Research map / Optimization / Implicit bias and generalization
Learning Compact Features via In-Training Representation Alignment
Xin Li, Xiangrui Li, Deng Pan, Yao Qiang, Dongxiao Zhu
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 |
|---|---|---|---|
| add class-conditional MMD alignment between two mini-batches | Resnet18 · KMNIST test | accuracy: 95.57 -> 95.79 | Improved |
| add class-conditional MMD alignment between two mini-batches | Resnet18 · FMNIST test | accuracy: 92.43 -> 92.57 | Improved |
| add class-conditional MMD alignment between two mini-batches | Resnet18 · CIFAR10 test | accuracy: 92.99 -> 93.70 | Improved |
| add class-conditional MMD alignment between two mini-batches | Resnet18 · STL10 test | accuracy: 70.88 -> 72.78 | Improved |
| add class-conditional MMD alignment between two mini-batches | Resnet18 · CIFAR100 test | accuracy: 74.19 -> 74.88 | Improved |
| add class-conditional MMD alignment between two mini-batches | Resnet-101 · CIFAR100 test | accuracy: 75.85 -> 76.85 | Improved |
| add class-conditional MMD alignment between two mini-batches | Resnet-18 · ImageNet test | accuracy: 50.01 -> 53.13 | Improved |
| add class-conditional MMD alignment between two mini-batches | Bert-base · AG's News test | accuracy: 92.10 -> 92.45 | Improved |
| add class-conditional MMD alignment between two mini-batches | Bert-base · Amazon Full test | accuracy: 58.53 -> 60.08 | Improved |
| add class-conditional MMD alignment between two mini-batches | Bert-base · Yahoo Answers test | accuracy: 66.60 -> 67.47 | Improved |
| add class-conditional MMD alignment between two mini-batches | Bert-base · Yelp Full test | accuracy: 61.12 -> 61.68 | Improved |
| replace vanilla SGD with center loss | Bert-base · AG's News test | accuracy: 92.10 -> 91.70 | Hurt |
Closest papers by meaning
- Tracing Representation Progression: Analyzing and Enhancing Layer-Wise Similarity (ICLR 2025 · similarity 0.86)
- Exploiting Invariance in Training Deep Neural Networks (AAAI 2022 · similarity 0.86)
- Gradient-Weight Alignment as a Train-Time Proxy for Generalization in Classification Tasks (NeurIPS 2025 · similarity 0.85)
- REPA Works Until It Doesn’t: Early-Stopped, Holistic Alignment Supercharges Diffusion Training (NeurIPS 2025 · similarity 0.85)
- Learning Input Encodings for Kernel-Optimal Implicit Neural Representations (ICML 2025 · similarity 0.85)
- Symmetric Neural-Collapse Representations with Supervised Contrastive Loss: The Impact of ReLU and Batching (ICLR 2024 · similarity 0.84)
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.
