Research map / Architectures and efficiency / CNN design
Inner Product-based Neural Network Similarity
Wei Chen, Zichen Miao, Qiang Qiu
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 |
|---|---|---|---|
| Select similar models by filter subspace similarity for ensemble | FedAvg · FL personalized ensemble | accuracy: 83.78 -> 85.82 | Improved |
| Select similar models by filter subspace similarity for ensemble | FedRep · FL personalized ensemble | accuracy: 76.44 -> 78.35 | Improved |
| Select similar models by filter subspace similarity for ensemble | Pretrain · FL personalized ensemble | accuracy: 81.77 -> 85.41 | Improved |
| Select ensemble members with filter subspace similarity instead of CKA | Pretrain · FL personalized ensemble | accuracy: 86.33 -> 85.41 | Hurt |
| Use filter subspace similarity instead of CCA for ensemble member selection | FedAvg · FL personalized ensemble | accuracy: 85.65 -> 85.82 | No clear effect |
| Replace CCA probing similarity with filter subspace similarity | ensemble selection · similarity computation cost | GFLOPs: 258,610 -> 0.019 | Improved |
| Replace CCA probing similarity with filter subspace similarity | ensemble selection · similarity computation cost | time (s): 1930.4 -> 0.016 | Improved |
| Add model ensemble selected by filter subspace similarity | AtomCL · 10-Split CIFAR-100 | accuracy: 78.11 -> 80.19 | Improved |
| Use filter subspace similarity instead of CKA for ensemble | AtomCL · 10-Split CIFAR-100 | accuracy: 80.01 -> 80.19 | No clear effect |
| Use CCA probing data from other datasets instead of CIFAR-100 | CCA similarity · CIFAR-100 same-dataset models | similarity: 0.8 -> 0.59 | Hurt |
| Use filter subspace similarity (probing-free) instead of CCA | same-dataset similarity · CIFAR-100 models | similarity: 0.59 -> 0.91 | Improved |
| Use proposed FL framework with shared coefficients and per-user atoms vs FedAvg | Ours · CIFAR-100 (100 clients, 5 classes) | accuracy: 82.39 -> 81.03 | Hurt |
Closest papers by meaning
- Tracing Representation Progression: Analyzing and Enhancing Layer-Wise Similarity (ICLR 2025 · similarity 0.87)
- Representational Similarity via Interpretable Visual Concepts (ICLR 2025 · similarity 0.86)
- Spectral Analysis of Representational Similarity with Limited Neurons (NeurIPS 2025 · similarity 0.86)
- Decoupling Semantic Similarity from Spatial Alignment for Neural Networks. (NeurIPS 2024 · similarity 0.85)
- How Not to Stitch Representations to Measure Similarity: Task Loss Matching Versus Direct Matching (AAAI 2025 · similarity 0.85)
- ReSi: A Comprehensive Benchmark for Representational Similarity Measures (ICLR 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.
