Research map / Architectures and efficiency / RNNs and normalization
Lifted Bregman Training of Neural Networks
Xiaoyu Wang, Martin Benning
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 lifted Bregman loss instead of classical quadratic-penalty lifted training | 4-layer MLP · MNIST (test) | accuracy: 86.3% -> 96.8% | Improved |
| use lifted Bregman loss instead of classical quadratic-penalty lifted training | 4-layer MLP · Fashion-MNIST (test) | accuracy: 80.0% -> 85.7% | Improved |
| train with lifted Bregman instead of SGD with back-propagation | 4-layer MLP · MNIST (test) | accuracy: 96.9% vs 96.8% | No clear effect |
| train with lifted Bregman instead of SGD with back-propagation | 4-layer MLP · Fashion-MNIST (test) | accuracy: 87.9% -> 85.7% | Hurt |
| replace quadratic penalty with Bregman penalty | Lifted network · MNIST hidden layers | percentage of linear nodes: 99.9% -> 47.4%, 82.2%, 70.1% | Improved |
| use lifted Bregman instead of GD/SGD with back-propagation | Sparse autoencoder · MNIST-1K | training objective value | Improved |
| use stochastic variant instead of deterministic | LBN-D / GD-BP · MNIST-1K sparse autoencoder | convergence speed | Improved |
| use lifted Bregman instead of GD/SGD with back-propagation | Sparse autoencoder · MNIST-1K | validation loss | Mixed |
| use lifted Bregman instead of GD/SGD with back-propagation | Sparse autoencoder · MNIST-1K | reconstruction quality | Improved |
| use lifted Bregman instead of GD/SGD with back-propagation | Sparse autoencoder · MNIST-1K | sparsity stability | Improved |
Closest papers by meaning
- Lifted Proximal Operator Machines (AAAI 2019 · similarity 0.88)
- A Bregman Proximal Viewpoint on Neural Operators (ICML 2025 · similarity 0.87)
- Better NTK Conditioning: A Free Lunch from (ReLU) Nonlinear Activation in Wide Neural Networks (NeurIPS 2025 · similarity 0.85)
- Random Sparse Lifts: Construction, Analysis and Convergence of finite sparse networks (ICLR 2024 · similarity 0.85)
- Convergence Analysis of Newton's Method for Neural Networks in the Overparameterized Limit (NeurIPS 2026 · similarity 0.84)
- Global Minimizers of $\ell^p$-Regularized Objectives Yield the Sparsest ReLU Neural Networks (NeurIPS 2025 · 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.
