Research map / Learning theory and algorithms / Combinatorial and search optimization
T-SKM-Net: Trainable Neural Network Framework for Linear Constraint Satisfaction via Sampling Kaczmarz-Motzkin Method
Haoyu Zhu, Yao Zhang, Jiashen Ren, Qingchun Hou
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 T-SKM-Net instead of Gurobi for L2 projection at large dimension | T-SKM-Net · L2 projection dim 10000 | time (s): 103.6 s -> 22.82 s | Improved |
| use T-SKM-Net instead of OptNet at 130 inequalities | T-SKM-Net · Constrained QP | time (ms): 450.6 ms -> 1.0 ms with similar objective (-10.25 vs -10.23) | Improved |
| use T-SKM-Net instead of DC3 at 150 inequalities | T-SKM-Net · Constrained QP | objective value: objective -9.06 -> -10.24, ineq violation 0.006 -> 0.000 | Improved |
| add T-SKM-Net layer to NN (post-processing) | NN · Constrained QP | max equality violation: equality violation 0.350 -> 0.000 | Improved |
| use T-SKM-Net instead of pandapower | T-SKM-Net · DCOPF IEEE 118-bus | time (ms): 141.08 ms -> 4.27 ms | Improved |
| use T-SKM-Net instead of DC3 | T-SKM-Net · DCOPF IEEE 118-bus | time (ms): 52.15 ms -> 4.27 ms; DC3 ineq violation 4e-4 | Improved |
| joint training instead of post-processing (ablation) | T-SKM-Net · S&P 500 portfolio allocation | Sharpe ratio: Sharpe 2.93 -> 3.02 | Improved |
| joint training instead of post-processing (ablation) | T-SKM-Net · DCOPF IEEE 118-bus | optimality gap / time: gap 1.0e-4 -> 7.0e-5; time 4.27 -> 5.64 ms | Mixed |
| add MSE punishment to upstream output in joint training (Train-p) (ablation) | T-SKM-Net · DCOPF IEEE 118-bus | optimality gap (%): gap 7.0e-5 -> 5.4e-5 | Improved |
| use T-SKM-Net instead of GLinSAT | T-SKM-Net · S&P 500 portfolio allocation | Sharpe ratio: Sharpe 2.28 -> 3.02 | Improved |
Closest papers by meaning
- AT-SKM-Net: An Accelerated Trainable Sampling Kaczmarz-Motzkin Framework for Linear Hard-Constraint Feasibility on Dynamic Graphs (NeurIPS 2026 · similarity 0.90)
- FSNet: Feasibility-Seeking Neural Network for Constrained Optimization with Guarantees (NeurIPS 2025 · similarity 0.85)
- Self-Supervised Transformers as Iterative Solution Improvers for Constraint Satisfaction (ICML 2025 · similarity 0.84)
- DeepSaDe: Learning Neural Networks That Guarantee Domain Constraint Satisfaction (AAAI 2024 · similarity 0.84)
- GLinSAT: The General Linear Satisfiability Neural Network Layer By Accelerated Gradient Descent (NeurIPS 2024 · similarity 0.84)
- Enforcing Hard Linear Constraints in Deep Learning Models with Decision Rules (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.
