Research map / Architectures and efficiency / RNNs and normalization
Abstraction Mechanisms Predict Generalization in Deep Neural Networks
Alex Gain, Hava Siegelmann
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 |
|---|---|---|---|
| Use CNA to predict test accuracy across architectures and datasets | MLP, VGG-18, ResNet-18, and ResNet-101 · ImageNet, CIFAR-10, CIFAR-100, SVHN, MNIST, and Fashion-MNIST | test accuracy | Improved |
| Use CNA-Margin rather than norm-based metrics to predict the generalization gap | DNNs across the evaluated architectures · Gaussian noise dataset | Pearson correlation with generalization gap | Improved |
| Use CNA-Margin rather than competing metrics to predict the generalization gap | DNNs across the evaluated architectures · Datasets with 10% to 50% shuffled training labels | Pearson correlation with generalization gap | Improved |
| Evaluate test error on inputs with higher information complexity | MLP · MNIST | test error: up to 16 times larger test error | Hurt |
| Train the MLP on MNIST and track CNA over training | MLP · MNIST | training loss | No clear effect |
Closest papers by meaning
- Explaining Generalization Power of a DNN Using Interactive Concepts (AAAI 2024 · similarity 0.88)
- Do Neural Networks Need Gradient Descent to Generalize? A Theoretical Study (NeurIPS 2025 · similarity 0.86)
- Understanding Generalization through Decision Pattern Shift (NeurIPS 2026 · similarity 0.86)
- Scaling can lead to compositional generalization (NeurIPS 2025 · similarity 0.86)
- Unveiling Memorization–Generalization Coexistence: A Case Study on Arithmetic Tasks with Label Noise (NeurIPS 2026 · similarity 0.85)
- Topological Generalization Bounds for Discrete-Time Stochastic Optimization Algorithms (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.
