Research map / Label-efficient and robust learning / Recognition and long-tail
Integrating Prior Knowledge in Contrastive Learning with Kernel
Benoit Dufumier, Carlo Alberto Barbano, Robin Louiset, Edouard Duchesnay, Pietro Gori
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 |
|---|---|---|---|
| Replace InfoNCE with Decoupled Uniformity | ResNet18 contrastive learning · CIFAR-10 | linear evaluation accuracy (%): 82.18_{\pm 0.30} -> \textbf{85.05}_{\pm 0.37} | Improved |
| Replace InfoNCE with Decoupled Uniformity | ResNet18 contrastive learning · CIFAR-100 | linear evaluation accuracy (%): 55.11_{\pm 0.20} -> \textbf{58.41}_{\pm 0.05} | Improved |
| Replace InfoNCE with Decoupled Uniformity | ResNet50 contrastive learning · ImageNet100 | linear evaluation accuracy (%): 68.76 -> 77.18 | Improved |
| Add BigBiGAN features as a kernel prior | Decoupled Uniformity · ImageNet100 | linear evaluation accuracy (%): 77.18 -> 78.02 | Improved |
| Use weak attributes as kernel prior rather than conditional positive sampling | Decoupled Uniformity versus CCLK · CUB | linear evaluation accuracy (%): 30.49 -> 32.87 | Improved |
| Use weak attributes as kernel prior rather than conditional positive sampling | Decoupled Uniformity versus CCLK · ImageNet100 | linear evaluation accuracy (%): 77.24 -> 79.02 | Improved |
| Use weak attributes as kernel prior rather than conditional positive sampling | Decoupled Uniformity versus CCLK · UT-Zappos | linear evaluation accuracy (%): 85.48 -> 85.56 | Improved |
| Add GloRIA features as a kernel prior | Decoupled Uniformity · CheXpert, Cardiomegaly | AUC (%): 86.39 -> 85.88 | Hurt |
| Add GloRIA features as a kernel prior | Decoupled Uniformity · CheXpert, Consolidation | AUC (%): 92.45 -> 93.03 | Improved |
| Add a VAE representation as kernel prior | Decoupled Uniformity · BIOBD, BD vs HC | linear evaluation AUC (%): 60.46_{\pm 1.23} -> \mathbf{62.19}_{\pm 1.58} | Improved |
| Use a VAE kernel prior when adding shared noisy bits to views | RandBits-CIFAR10 representation learning · RandBits-CIFAR10, 10 bits | linear evaluation accuracy (%): 13.67 -> \textbf{68.42}_{\pm 0.51} | Improved |
| Use a VAE kernel prior when adding shared noisy bits to views | RandBits-CIFAR10 representation learning · RandBits-CIFAR10, 20 bits | linear evaluation accuracy (%): 10.07 -> \textbf{68.58}_{\pm 0.17} | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- GLoRIA — GLoRIA: A Multimodal Global-Local Representation Learning Framework for Label-Efficient Medical Image Recognition (ICCV 2021) · better on 1 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- A Statistical Theory of Contrastive Learning via Approximate Sufficient Statistics (NeurIPS 2025 · similarity 0.88)
- On the Similarities of Embeddings in Contrastive Learning (ICML 2025 · similarity 0.88)
- Contrastive Learning Can Find An Optimal Basis For Approximately View-Invariant Functions (ICLR 2023 · similarity 0.88)
- Diversity Is All You Need for Contrastive Learning: Spectral Bounds on Gradient Magnitudes (NeurIPS 2025 · similarity 0.87)
- Variational Supervised Contrastive Learning (NeurIPS 2025 · similarity 0.87)
- Bridging Mini-Batch and Asymptotic Analysis in Contrastive Learning: From InfoNCE to Kernel-Based Losses (ICML 2024 · similarity 0.87)
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.
