Research map / Architectures and efficiency / RNNs and normalization
Revisiting Model Stitching to Compare Neural Representations
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| stitch self-supervised ResNet-50 bottom layers into supervised network | ResNet-50 ImageNet · ImageNet | stitching penalty | No clear effect |
| stitch SimCLR ResNet-18 into supervised ResNet-18 on CIFAR-10 | ResNet-18 · CIFAR-10 | stitching penalty: max 3% | No clear effect |
| stitch bottom layers from networks with different random init | CNNs/ResNets/ViT · CIFAR-10 | test error: up to 3% error | No clear effect |
| plug in representation trained with more samples (25K vs 10K) | ResNet-18 · CIFAR-10 | test error | Helped |
| plug in wider-network representation into thinner network | ResNet-18 · CIFAR-10 | test error | Helped |
| plug in longer-trained representation | ResNet-18 · CIFAR-10 | test error | Helped |
| stitch bottom layers trained with 100% noisy labels (ablation) | ResNet-18 · CIFAR-10 | stitching penalty | Hurt |
| stitch bottom layers trained with coarse or 10-50% noisy labels (ablation) | ResNet-18 · CIFAR-10 | stitching penalty | No clear effect |
Closest papers by meaning
- Functional Alignment Can Mislead: Examining Model Stitching (ICML 2025 · similarity 0.89)
- How Not to Stitch Representations to Measure Similarity: Task Loss Matching Versus Direct Matching (AAAI 2025 · similarity 0.89)
- Towards a learning theory of representation alignment (ICLR 2025 · similarity 0.89)
- Latent Functional Maps: a spectral framework for representation alignment (NeurIPS 2024 · similarity 0.87)
- Connecting Neural Models Latent Geometries with Relative Geodesic Representations (NeurIPS 2025 · similarity 0.86)
- From Bricks to Bridges: Product of Invariances to Enhance Latent Space Communication (ICLR 2024 · similarity 0.86)
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.