Research map / Architectures and efficiency / RNNs and normalization
Simplicity Bias in 1-Hidden Layer Neural Networks
Depen Morwani, jatin batra, Prateek Jain, Praneeth Netrapalli
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 |
|---|---|---|---|
| Train a second model on features orthogonal to the first model's important directions (OrthoP), rather than training it independently | ResNet-50 feature-based 1-hidden-layer classifiers · B-Imagenette, Imagenette, Waterbirds, MNIST-CIFAR | Mistake diversity: 3.87± 1.54 -> 21.15±1.57; 6.6±0.46 -> 11.44±0.65; 2.9±0.52 -> 14.53±0.48; 0.0±0.0 -> 5.56±7.89 | Improved |
| Train a second model on features orthogonal to the first model's important directions (OrthoP), rather than training it independently | ResNet-50 feature-based 1-hidden-layer classifiers · B-Imagenette, Imagenette, Waterbirds, MNIST-CIFAR | Class-conditioned logit correlation: 99.88±0.01 -> 90.86±1.08; 99.31±0.12 -> 91±0.59; 99.66±0.04 -> 93.81±0.48; 99.76±0.17 -> 78.74±2.28 | Improved |
| Use a low-dimensional projection to isolate the trained model's influential input directions | 1-hidden-layer FCNs in the rich regime · b-Imagenette, Imagenette, Waterbirds, MNIST-CIFAR | Randomized accuracy under P versus P_perp mixing: P_perp-RA: 89.94±0.22, 75.89±0.25, 91.47±0.11, 94.15±0.21; P-RA: 49.53±0.24, 9.33±0.01, 62.51±0.07, 55.2±0.13 | Mixed |
| Find a low-dimensional influential projection despite high-rank first-layer weights | 1-hidden-layer FCNs in the lazy regime · b-Imagenette, Imagenette, Waterbirds, MNIST-CIFAR | Randomized accuracy under P versus P_perp mixing: P_perp-RA: 90.07±0.34, 68.25±1.18, 89.67±0.42, 99.45±0.17; P-RA: 52.09±1.34, 11.92±0.82, 62.44±4.48, 49.83±0.67 | Mixed |
| Train the second ensemble model on P_perp-projected features and average its logits with the first model | OrthoP ensemble · Binary-Imagenette, Imagenette, Waterbirds-Landbirds under Gaussian noise | Test accuracy | Improved |
Closest papers by meaning
- Simplicity Bias of Two-Layer Networks beyond Linearly Separable Data (ICML 2024 · similarity 0.92)
- Do We Always Need the Simplicity Bias? Looking for Optimal Inductive Biases in the Wild (CVPR 2025 · similarity 0.89)
- Simplicity Bias via Global Convergence of Sharpness Minimization (ICML 2024 · similarity 0.88)
- Simplicity Bias in Overparameterized Machine Learning (AAAI 2024 · similarity 0.87)
- The Rich and the Simple: On the Implicit Bias of Adam and SGD (NeurIPS 2025 · similarity 0.87)
- Neural Networks Learn Statistics of Increasing Complexity (ICML 2024 · similarity 0.87)
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.
