Research map / Optimization / Implicit bias and generalization
Deep-ICE: The first globally optimal algorithm for empirical risk minimization of two-layer maxout and ReLU networks
Xi He, Yi Miao, Max A Little
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 |
|---|---|---|---|
| exact 0-1 loss optimization (Deep-ICE) instead of gradient descent | rank-2 maxout · VP dataset N=704 | training 0-1 loss: 25 -> 16 | Improved |
| Deep-ICE K=2 with coresets instead of SVM | two-layer maxout · UCI Caesarian | test accuracy (%): 57.33 -> 88.00 | Improved |
| Deep-ICE K=2 with coresets instead of SVM | two-layer maxout · UCI DB | test accuracy (%): 67.62 -> 81.37 | Improved |
| Deep-ICE K=2 instead of MLP K=2 trained by Adam | two-layer maxout · UCI SS | test accuracy (%): 79.65 -> 86.72 | Improved |
| Deep-ICE K=2 instead of EXACT K=2 | two-layer maxout · UCI Voicepath | test accuracy (%): 92.47 -> 97.59 | Improved |
| Deep-ICE K=2 instead of EXACT K=2 | two-layer maxout · UCI Ai4i | test accuracy (%): 96.60 -> 97.82 | Improved |
| increase hidden units from K=1 to K=2 (ablation) | Deep-ICE · UCI Caesarian | test accuracy (%): 82.67 -> 88.00 | Improved |
| increase hidden units from K=2 to K=3 (ablation) | Deep-ICE · UCI Caesarian | test accuracy (%): 88.00 -> 86.67 | Hurt |
| increase hidden units from K=2 to K=3 (ablation) | Deep-ICE · UCI SO | test accuracy (%): 75.33 -> 74.95 | No clear effect |
| Deep-ICE exact solutions generalize despite higher training accuracy | Deep-ICE · UCI datasets | test accuracy | Improved |
Closest papers by meaning
- Simplicity Bias and Optimization Threshold in Two-Layer ReLU Networks (ICML 2025 · similarity 0.87)
- Optimal Rates for Generalization of Gradient Descent for Deep ReLU Classification (NeurIPS 2025 · similarity 0.86)
- Benign Overfitting and Grokking in ReLU Networks for XOR Cluster Data (ICLR 2024 · similarity 0.86)
- Large Stepsize Gradient Descent for Non-Homogeneous Two-Layer Networks: Margin Improvement and Fast Optimization (NeurIPS 2024 · similarity 0.86)
- Global Minimizers of $\ell^p$-Regularized Objectives Yield the Sparsest ReLU Neural Networks (NeurIPS 2025 · similarity 0.86)
- On the Hardness of Training Deep Neural Networks Discretely (AAAI 2025 · similarity 0.86)
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.
