Research map / Label-efficient and robust learning / Recognition and long-tail
Stretching Beyond the Obvious: A Gradient-Free Framework to Unveil the Hidden Landscape of Visual Invariance
Lorenzo Tausani, Paolo Muratore, Morgan Bruce Talbot, Giacomo Amerio, Gabriel Kreiman, Davide Zoccolan
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 |
|---|---|---|---|
| optimize with both stretch and squeeze losses instead of one (ablation) | SnS · invariant/adversarial image synthesis | image properties | Improved |
| use SnS invariant images instead of affine transformations of the MEI | SnS · L2-robust ResNet50 readout units | pixel distance vs activation preserved: L2 distance 271 with 34% activation drop | Improved |
| stretch deeper layer representations instead of pixel space (robust network) (ablation) | SnS on L2-robust ResNet50 · human 12-AFC recognition | human classification accuracy | Hurt |
| stretch deeper layer representations instead of pixel space (standard network) (ablation) | SnS on standard ResNet50 · human 12-AFC recognition | human classification accuracy | Improved |
| L2 adversarial training of the subject network | ResNet50 · human recognition of SnS invariant images | human classification accuracy | Improved |
| L2 robustification effect at deeper stretching layers | ResNet50 · interpretability gap robust vs standard | human classification accuracy | Mixed |
| use L-infinity instead of L2 robustification | ResNet50 · observer recognition of invariant images | observer accuracy | Improved |
| subsample the stretched layer representation (ablation) | SnS · representational distance of invariant images | qualitative / distance | No clear effect |
Closest papers by meaning
- Diverse feature visualizations reveal invariances in early layers of deep neural networks (ECCV 2018 · similarity 0.90)
- Finding Biological Plausibility for Adversarially Robust Features via Metameric Tasks (ICLR 2022 · similarity 0.87)
- From Bricks to Bridges: Product of Invariances to Enhance Latent Space Communication (ICLR 2024 · similarity 0.86)
- Discriminating image representations with principal distortions (ICLR 2025 · similarity 0.86)
- In-Context Symmetries: Self-Supervised Learning through Contextual World Models (NeurIPS 2024 · similarity 0.85)
- Bridging Critical Gaps in Convergent Learning: How Representational Alignment Evolves Across Layers, Training, and Distribution Shifts (NeurIPS 2025 · similarity 0.85)
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.
