Research map / Learning theory and algorithms / Optimal transport
Unbalanced Soft-Matching Distance For Neural Representational Comparison With Partial Unit Correspondence
Chaitanya Kapoor, Alex H Williams, Meenakshi Khosla
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 |
|---|---|---|---|
| allow partial matching (partial soft-matching) instead of full soft-matching | soft-matching distance · NSD voxel alignment precision (V1d + V2v) | precision: 0.881 -> 0.971 | Improved |
| use partial soft-matching instead of soft-matching | soft-matching distance · NSD voxel alignment precision, 15 region pairs | precision: V1v + V1d 0.839 -> 0.905; V2d + V3v 0.833 -> 0.971 | Improved |
| partial soft-matching versus noise-ceiling thresholding eps=0.3 on V1v + V2v | ParSM · NSD voxel alignment precision (V1v + V2v) | precision: 0.695 (eps=0.3) vs 0.680 | Hurt |
| rank units with partial soft-matching versus correlation-based ordering | alignment-based unit removal · ResNet-18 and NSD | alignment score | Improved |
| single partial soft-matching optimization instead of brute-force greedy removal | unit ranking · ResNet-18 and NSD | alignment score | No clear effect |
| use partial soft-matching with L-curve in the presence of spurious neurons | soft-matching · Synthetic signal+noise neurons | transport cost / correct matches | Improved |
| apply random rotation to one network's representation | ResNet-18 matched subpopulations · ImageNet ResNet-18 alignment | alignment | Hurt |
| use squared Euclidean cost instead of cosine (ablation) | partial soft-matching · synthetic, ResNet-18, NSD | alignment | No clear effect |
Closest papers by meaning
- Bridging Critical Gaps in Convergent Learning: How Representational Alignment Evolves Across Layers, Training, and Distribution Shifts (NeurIPS 2025 · similarity 0.87)
- How Not to Stitch Representations to Measure Similarity: Task Loss Matching Versus Direct Matching (AAAI 2025 · similarity 0.85)
- Sparse components distinguish visual pathways & their alignment to neural networks (ICLR 2025 · similarity 0.85)
- Unbalancedness in Neural Monge Maps Improves Unpaired Domain Translation (ICLR 2024 · similarity 0.84)
- Scaling Laws for Task-Optimized Models of the Primate Visual Ventral Stream (ICML 2025 · similarity 0.84)
- Spectral Analysis of Representational Similarity with Limited Neurons (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.
