Research map / Architectures and efficiency / RNNs and normalization
Towards Disentangling Information Paths with Coded ResNeXt
Apostolos Avranas, Marios Kountouris
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 |
|---|---|---|---|
| force subNN specialization with coding (Coded ResNeXt vs ResNeXt) | ResNeXt · CIFAR-10 | accuracy: 93.66% -> 94.41% | Improved |
| same on CIFAR-100 | ResNeXt · CIFAR-100 | accuracy: 76.86% -> 78.76% | Improved |
| same on ImageNet | ResNeXt-50 · ImageNet | accuracy: 79.50% -> 80.24% | Improved |
| remove inactive subNNs from a block (ablation) | Coded ResNeXt · CIFAR/ImageNet val | accuracy | Improved |
| remove active subNNs from a block (ablation) | Coded ResNeXt · CIFAR/ImageNet val | accuracy | Hurt |
| extract binary classifiers from ResNeXt instead of Coded ResNeXt | ResNeXt · CIFAR-10 | precision: 93% -> 13% | Hurt |
| concatenate subNN outputs instead of summing (ablation) | Coded ResNeXt · BC extraction | precision: precision < 20% | Hurt |
| increase mu from 2 to 4 (ablation) | Coded ResNeXt-50 · ImageNet | accuracy: 80.24% -> 79.81% | Hurt |
| increase mu from 2 to 4 (ablation) | Coded ResNeXt-50 · ImageNet | early decoder accuracy: 2.38% -> 4.6% | Improved |
| use stage ratios (24/32,16/32,8/32) at s2-s4 instead of (16/32, 8/32) at s3,s4 (ablation) | Coded ResNeXt-50 · ImageNet 150 epochs | accuracy: 78.1% -> worse than 77% | Hurt |
| add early-decoder agreement to temperature scaling (h_aug vs h_T) | Calibration · CIFAR-10 | MCE: 24.35% -> 24.14% | Improved |
| add early-decoder agreement to temperature scaling | Calibration · CIFAR-10 | ECE: 0.70% -> 0.70% | No clear effect |
Closest papers by meaning
- Codebook Features: Sparse and Discrete Interpretability for Neural Networks (ICML 2024 · similarity 0.85)
- Residual Alignment: Uncovering the Mechanisms of Residual Networks (NeurIPS 2023 · similarity 0.85)
- Layer Compression of Deep Networks with Straight Flows (AAAI 2024 · similarity 0.85)
- PICNN: A Pathway towards Interpretable Convolutional Neural Networks (AAAI 2024 · similarity 0.84)
- $\mu$PC: Scaling Predictive Coding to 100+ Layer Networks (NeurIPS 2025 · similarity 0.84)
- Masked Completion via Structured Diffusion with White-Box Transformers (ICLR 2024 · 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.
