Research map / Label-efficient and robust learning / Recognition and long-tail
GULP: a prediction-based metric between representations
Enric Boix-Adserà, Hannah Lawrence, George Stepaniants, Philippe Rigollet
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 |
|---|---|---|---|
| Measure representation distance using uniform prediction differences over regularized linear tasks | ImageNet pretrained representations · ImageNet; linear regression | Spearman's ρ with prediction differences | Improved |
| Use GULP with larger regularization | ImageNet pretrained representations · ImageNet architecture clustering | architecture separation | Improved |
| Increase GULP regularization | Independently trained ResNet18 representations · CIFAR10 training | distance trend across training | Improved |
| Use GULP to predict generalization differences for logistic rather than linear prediction | GULP · Downstream logistic prediction | prediction-gap correlation | Hurt |
| Set λ=0, recovering the CCA-based distance (ablation) | GULP · Representation comparison | distance invariance | Mixed |
| Set λ=10⁻⁶ rather than using larger regularization (ablation) | GULP on MNIST network representations · MNIST train | architecture clustering | Mixed |
| Use GULP with λ=1 rather than λ=10⁻⁶ (ablation) | GULP on MNIST network representations · MNIST train | architecture clustering | Improved |
| Use small λ rather than large λ during training (ablation) | GULP on independently trained ResNet18 representations · CIFAR10 training | sensitivity to representation differences | Mixed |
Closest papers by meaning
- ReSi: A Comprehensive Benchmark for Representational Similarity Measures (ICLR 2025 · similarity 0.85)
- Tracing Representation Progression: Analyzing and Enhancing Layer-Wise Similarity (ICLR 2025 · similarity 0.85)
- How Not to Stitch Representations to Measure Similarity: Task Loss Matching Versus Direct Matching (AAAI 2025 · similarity 0.84)
- The Unreasonable Effectiveness of Linear Prediction as a Perceptual Metric (ICLR 2024 · similarity 0.84)
- Latent Functional Maps: a spectral framework for representation alignment (NeurIPS 2024 · similarity 0.84)
- From Flat to Hierarchical: Extracting Sparse Representations with Matching Pursuit (NeurIPS 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.
