Research map / Label-efficient and robust learning / Self-supervised and contrastive
Okapi: Generalising Better by Making Statistical Matches Match
Myles Bartlett, Sara Romiti, Viktoriia Sharmanska, Novi Quadrianto
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 |
|---|---|---|---|
| switch backbone from ResNet to ConvNeXt | ERM · iWildCam | OOD macro F1: 32.2 -> 33.3 | Improved |
| switch backbone from ResNet to ConvNeXt | FixMatch · iWildCam | OOD macro F1: 31.0 -> 35.2 | Improved |
| switch backbone from ResNet to ConvNeXt | ERM · PovertyMap | OOD worst U/R correlation: 0.49 -> 0.53 | Improved |
| switch backbone from ResNet to ConvNeXt | FixMatch · PovertyMap | OOD worst U/R correlation: 0.30 -> 0.34 | Improved |
| use ConvNeXt so FixMatch surpasses ERM | FixMatch · iWildCam | OOD macro F1: 31.0 vs 32.2 -> 35.2 vs 33.3 | Improved |
| add Okapi cross-domain consistency loss | ERM (ConvNeXt) · iWildCam | OOD macro F1: 33.3 -> 36.1 | Improved |
| add Okapi | FixMatch (ConvNeXt) · iWildCam | OOD macro F1: 35.2 -> 36.1 | Improved |
| add Okapi | ERM (ConvNeXt) · PovertyMap | OOD worst U/R correlation: 0.53 -> 0.55 | Improved |
| add Okapi | ERM (ConvNeXt) · PovertyMap | OOD worst U/R MSE: 0.35 -> 0.33 | No clear effect |
| FixMatch without confidence thresholding for regression | FixMatch · PovertyMap | OOD worst U/R correlation: 0.53 -> 0.34 | Hurt |
| add Okapi | ERM (reproduction) · CivilComments | OOD worst-group accuracy: 68.5 -> 69.7 | Improved |
| remove caliper filtering (standard kNN) (ablation) | Okapi · PovertyMap | OOD worst U/R correlation: 0.54 -> 0.55 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- FixMatch [63] — FixMatch: Simplifying Semi-Supervised Learning with Consistency and Confidence (NeurIPS 2020) · better on 9 of 10 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
- ECACL: A Holistic Framework for Semi-Supervised Domain Adaptation (ICCV 2021 · similarity 0.86)
- Open-World Semi-Supervised Learning (ICLR 2022 · similarity 0.86)
- FixMatch: Simplifying Semi-Supervised Learning with Consistency and Confidence (NeurIPS 2020 · similarity 0.86)
- Trash To Treasure: Harvesting OOD Data With Cross-Modal Matching for Open-Set Semi-Supervised Learning (ICCV 2021 · similarity 0.86)
- Scaling Up Semi-supervised Learning with Unconstrained Unlabelled Data (AAAI 2024 · similarity 0.85)
- RankUp: Boosting Semi-Supervised Regression with an Auxiliary Ranking Classifier (NeurIPS 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.
