Research map / Probabilistic and scientific modeling / Causal discovery
Readout Representation: Redefining Neural Codes by Input Recovery
Shunsuke Onoo, Yoshihiro Nagano, Yukiyasu Kamitani
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 |
|---|---|---|---|
| perturb VGG19 lower-layer features to 0.7 correlation distance before inversion (ablation) | feature inversion readout · VGG19 lower layers | pixel-space correlation distance of reconstruction: 0.7 feature perturbation still within 0.1 pixel distance | No clear effect |
| use features from higher layers instead of lower-to-middle layers (ablation) | feature inversion from perturbed features · VGG19, DINOv2, CLIP | recovery under perturbation | Hurt |
| use GPT2 (and some OPT variants) instead of BERT/OPT-350m (ablation) | text feature inversion · C4 texts | token error rate under perturbation | Hurt |
| compare correctly classified vs misclassified images | representation size · VGG19, ImageNet val (8 vs 8 images) | representation size | Improved |
| use random uniform noise images instead of natural images | representation size · VGG19 | representation size: zero across all layers | Hurt |
| increase feature dimensionality of layer | representation size · vision models across layers | representation size | Mixed |
| remove DIP and invert in pixel space (ablation) | representation size · VGG19 | representation size | Hurt |
Closest papers by meaning
- Decomposing stimulus-specific sensory neural information via diffusion models (NeurIPS 2025 · similarity 0.86)
- Not all solutions are created equal: An analytical dissociation of functional and representational similarity in deep linear neural networks (ICML 2025 · similarity 0.86)
- Unlearning-based Neural Interpretations (ICLR 2025 · similarity 0.85)
- From Flat to Hierarchical: Extracting Sparse Representations with Matching Pursuit (NeurIPS 2025 · similarity 0.85)
- The Non-Linear Representation Dilemma: Is Causal Abstraction Enough for Mechanistic Interpretability? (NeurIPS 2025 · similarity 0.85)
- How Learning by Reconstruction Produces Uninformative Features For Perception (ICML 2024 · 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.
