Research map / Label-efficient and robust learning / Self-supervised and contrastive
RELAX: Representation Learning Explainability
Kristoffer Wickstrøm, Daniel J. Trosten, Sigurd Løkse, Ahcène Boubekki, Karl Øyvind Mikalsen, Michael Kampffmeyer, Robert Jenssen
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 RELAX instead of Saliency | supervised ResNet · Pointing game (VOC) | pointing game: 67.1 -> 72.6 | Improved |
| use RELAX instead of Saliency | SimCLR · Pointing game (COCO) | pointing game: 75.9 -> 85.2 | Improved |
| use RELAX instead of Saliency | SwAV · Relevance rank (VOC) | relevance rank: 40.9 -> 52.4 | Improved |
| add uncertainty filtering (U-RELAX) to RELAX (ablation) | supervised features · Monotonicity | monotonicity: 18.3 -> 23.6 | Improved |
| add uncertainty filtering (U-RELAX) to RELAX (ablation) | SwAV features · Monotonicity | monotonicity: 21.3 -> 18.3 | Hurt |
| add uncertainty filtering (U-RELAX) to RELAX (ablation) | SimCLR · Pointing game (VOC) | pointing game: 68.7 -> 68.6 | No clear effect |
| add uncertainty filtering (U-RELAX) to RELAX (ablation) | supervised features · Relevance rank (VOC) | relevance rank: 56.4 -> 52.4 | Hurt |
| use gamma 1.0 instead of 0.95 in U-RELAX (median) (ablation) | U-RELAX · Monotonicity, supervised | monotonicity: 17.9 -> 23.6 | Improved |
| use median instead of mean aggregation in U-RELAX (gamma 1.0) (ablation) | U-RELAX · Monotonicity, supervised | monotonicity: 23.2 -> 23.6 | Improved |
| mask with zeros instead of Gaussian noise (ablation) | RELAX · Pointing game (VOC), SimCLR | pointing game: 66.6 -> 68.7 | Improved |
| mask with zeros instead of Gaussian noise (ablation) | RELAX · Pointing game (COCO), SwAV | pointing game: 85.1 -> 84.7 | No clear effect |
Later papers that compared with this method
1 later accepted paper reports RELAX as a baseline in their results tables; 0 of them report a better number on most of the results they share with it.
- Finding NEM-U: Explaining unsupervised representation learning through neural network generated explanation masks (ICML 2024 · better on 1 of 9 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
- Representation Learning: A Causal Perspective (AAAI 2025 · similarity 0.85)
- From Causal to Concept-Based Representation Learning (NeurIPS 2024 · similarity 0.85)
- Learning Linear Causal Representations from General Environments: Identifiability and Intrinsic Ambiguity (NeurIPS 2024 · similarity 0.85)
- The Non-Linear Representation Dilemma: Is Causal Abstraction Enough for Mechanistic Interpretability? (NeurIPS 2025 · similarity 0.84)
- Towards the Causal Complete Cause of Multi-Modal Representation Learning (ICML 2025 · similarity 0.84)
- Contextures: Representations from Contexts (ICML 2025 · 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.
