Research map / Label-efficient and robust learning / Self-supervised and contrastive
Self-Supervised Representation Learning From Multi-Domain Data
Zeyu Feng, Chang Xu, Dacheng Tao
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 |
|---|---|---|---|
| add MI constraints to naive multi-domain combination (DeepAll) | RotNet on AlexNet · PACS average, conv5 and fc7 | top-1 accuracy: 66.2 -> 67.3 (conv5), 59.3 -> 64.5 (fc7) | Improved |
| add MI constraints to naive combination with AET | AET on AlexNet · PACS average, conv5 and fc7 | top-1 accuracy: 66.1 -> 68.8 (conv5), 52.1 -> 59.3 (fc7) | Improved |
| train on all four PACS domains naively instead of a single domain | RotNet · PACS average, fc7 | top-1 accuracy: 59.3 vs 58.0 (Art), 61.5 (Cartoon) | Mixed |
| add Places to ImageNet naively for pretraining | RotNet · ImageNet val, conv5 | top-1 accuracy: 31.9 -> 31.6 | Hurt |
| add MI constraints to ImageNet+Places combination | RotNet · ImageNet and Places average, fc7 | top-1 accuracy: 25.0 -> 28.9 | Improved |
| add MI constraints to ImageNet+Places pretraining | RotNet · PASCAL VOC 2007, fc7 | mAP: 73.8 -> 75.6 | Improved |
| add PASCAL training images to ImageNet pretraining | RotNet · PASCAL VOC 2007, conv5 | mAP: 74.3 -> 74.8 | Improved |
| add MI constraints to ImageNet+PASCAL pretraining | RotNet · PASCAL VOC 2007 | mAP: 74.8 -> 74.8 (conv5), 73.2 -> 75.3 (fc7) | Mixed |
| add domain-invariance constraint only (ablation) | DeepAll RotNet · PACS average, conv5 and fc7 | top-1 accuracy: 66.2 -> 67.0 (conv5), 59.3 -> 59.1 (fc7) | Mixed |
| add domain-specific constraint only (ablation) | DeepAll RotNet · PACS average, conv5 and fc7 | top-1 accuracy: 59.3 -> 62.8 (fc7), 66.2 -> 66.3 (conv5) | Improved |
| combine both constraints (lambda_u=0.1, lambda_l=0.1) (ablation) | DeepAll RotNet · PACS average, conv5 and fc7 | top-1 accuracy: 66.2 -> 67.2 (conv5), 59.3 -> 62.5 (fc7) | Improved |
| raise lambda_l from 0.1 to 1 with lambda_u 0.1 (ablation) | Full model · PACS average, fc7 | top-1 accuracy: 62.5 -> 64.5 | Improved |
Closest papers by meaning
- Self-Supervised Learning Across Domains (TPAMI 2021 · similarity 0.88)
- Visual Representation Learning over Latent Domains (ICLR 2022 · similarity 0.87)
- Learning Representations by Maximizing Mutual Information Across Views (NeurIPS 2019 · similarity 0.87)
- Self-supervised Learning from a Multi-view Perspective (ICLR 2021 · similarity 0.86)
- Self-Supervised Learning from Structural Invariance (ICLR 2026 · similarity 0.86)
- A General Representation-Based Approach to Multi-Source Domain Adaptation (ICML 2025 · similarity 0.85)
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.
