Research map / Label-efficient and robust learning / Self-supervised and contrastive
Difficulty-Based Sampling for Debiased Contrastive Representation Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 relative-difficulty weighting to train the debiased encoder instead of DCL | WCL · CIFAR-10 | ACC (worst): 81.2±0.2 -> 83.3±0.8 | Helped |
| Use relative-difficulty weighting to train the debiased encoder instead of HCL | WCL · CIFAR-100 | ACC (worst): 29.00 ±0.8 -> 31.00±0.7 | Helped |
| Use relative-difficulty weighting to train the debiased encoder instead of HCL | WCL · Waterbirds | ACC (worst): 5.26 ± 1.1 -> 31.58 ± 3.5 | Helped |
| Use relative-difficulty weighting to train the debiased encoder instead of HCL | WCL · CelebA | ACC (worst): 52.13 ± 0.8 -> 57.40 ± 1.2 | Helped |
| Use WCL rather than the supervised baseline | WCL · Adult | EOd: 0.185 -> 0.104 | Helped |
| Use WCL rather than the supervised baseline | WCL · COMPAS | EOd: 0.124 -> 0.113 | Helped |
| Use the triplet-trained biased encoder instead of the debiased encoder | WCL · CIFAR-100 | ACC (worst): 31.00±0.7 -> 1.0 ± 0.5 | Hurt |
| Use WCL rather than SimCLR | WCL · Waterbirds | ACC (top-1): 77.80 ± 1.5 -> 76.92 ± 0.3 | No clear effect |
Closest papers by meaning
- Debiased Contrastive Learning (NeurIPS 2020 · similarity 0.92)
- Contrastive Learning with Hard Negative Samples (ICLR 2021 · similarity 0.92)
- On the Surrogate Gap between Contrastive and Supervised Losses (ICML 2022 · similarity 0.89)
- $\mathbb{X}$-Sample Contrastive Loss: Improving Contrastive Learning with Sample Similarity Graphs (ICLR 2025 · similarity 0.87)
- Self-Supervised Contrastive Learning is Approximately Supervised Contrastive Learning (NeurIPS 2025 · similarity 0.87)
- On the Similarities of Embeddings in Contrastive Learning (ICML 2025 · 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.