Research map / Learning theory and algorithms / Sample complexity
Logit Distance Bounds Representational Similarity
Beatrix Miranda Ginn Nielsen, Emanuele Marconato, Luigi Gresele, Andrea Dittadi, Simon Buchholz
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 |
|---|---|---|---|
| replace KL distillation loss with L1 logit distance (ablation) | student · Synth | d_rep: 49.5 -> 1.44 | Improved |
| replace KL distillation loss with squared logit distance (ablation) | student · Synth | d_rep: 49.5 -> 0.20 | Improved |
| replace KL loss with L1 logit distance (ablation) | student · Synth | mCCA: 0.580 -> 0.999 | Improved |
| replace KL loss with L1 logit distance (ablation) | student · CIFAR-100 | d_rep: 87.6 -> 4.56 | Improved |
| replace KL loss with squared logit distance (ablation) | student · CIFAR-100 | d_rep: 87.6 -> 0.66 | Improved |
| replace KL loss with L1 logit distance (ablation) | student · CIFAR-100 | mCCA: 0.515 -> 0.767 | Improved |
| replace KL loss with L1 logit distance (ablation) | student · CIFAR-100 | Acc(Y): 0.480 -> 0.492 | Improved |
| replace KL loss with L1 logit distance (ablation) | student · SUB | Acc(C): 0.06 -> 0.75 | Improved |
| replace KL loss with squared logit distance (ablation) | student · SUB | Acc(C): 0.06 -> 0.72 | Improved |
| replace KL loss with L1 logit distance (ablation) | student · SUB | d_rep: 3100 -> 10.0 | Improved |
| replace KL loss with squared logit distance (ablation) | student · SUB | d_rep: 3100 -> 1.6 | Improved |
| replace KL loss with L1 logit distance (ablation) | student · SUB | mCCA: 0.42 -> 0.99 | Improved |
Closest papers by meaning
- When Does Closeness in Distribution Imply Representational Similarity? An Identifiability Perspective (NeurIPS 2025 · similarity 0.92)
- Objective drives the consistency of representational similarity across datasets (ICML 2025 · similarity 0.87)
- Nonparametric Identification of Latent Concepts (ICML 2025 · similarity 0.86)
- Scaled Decoupled Distillation (CVPR 2024 · similarity 0.85)
- Towards the Causal Complete Cause of Multi-Modal Representation Learning (ICML 2025 · similarity 0.85)
- Learning Linear Causal Representations from General Environments: Identifiability and Intrinsic Ambiguity (NeurIPS 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.
