Research map / Learning theory and algorithms / Combinatorial and search optimization
Fast Computation of Superquantile-Constrained Optimization Through Implicit Scenario Reduction
Jake Roth, Ying Cui
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 ALM instead of Gurobi at high risk aversion (k=1%m) with the largest m | superquantile-constrained problems · Synthetic, m up to 2^20 | solve time: at least 20x and up to 70x | Improved |
| use ALM instead of Gurobi at lower risk aversion (larger k) | superquantile-constrained problems · Synthetic, large-m instances | solve time: at least 5x and up to 30x | Improved |
| use ALM instead of Gurobi dual simplex for small m and large n with linear objective | superquantile-constrained problems · Synthetic, small-m large-n, linear objective | solve time: GRB beats ALM by up to 4x | Hurt |
| use ALM instead of outer approximation (G-OA) in high risk-aversion instances | superquantile-constrained problems · Synthetic, large-m, k=1%m | solve time: about 5-10x | Improved |
| use ALM (eps=1e-3) instead of OSQP (eps=1e-3) | superquantile-constrained problems · Synthetic, large-m instances | solve time: at least 50x, sometimes more than 750x | Improved |
| increase scenario count m or number of constraints L (ablation) | ALM relative to Gurobi · Synthetic, k=1%m | relative solve time: ALM never slower than GRB; 20x to 70x at largest m | Improved |
| use larger k (lower risk aversion) in ALM (ablation) | ALM · Synthetic, large-m instances | relative performance versus Gurobi: advantage drops from 20-70x to 5-30x | Hurt |
| use second-order (Jacobian) information versus first-order in ALM when m much larger than n and k much smaller than m (ablation) | ALM subproblem · Synthetic ten largest-m instances | incremental cost: 3-7x cheaper | Improved |
| use PSG instead of ALM | PSG HELI solver · Synthetic, large-m instances | primal feasibility | Hurt |
Closest papers by meaning
- Corporate Needs You to Find the Difference: Revisiting Submodular and Supermodular Ratio Optimization Problems (NeurIPS 2025 · similarity 0.85)
- BayeSQP: Bayesian Optimization through Sequential Quadratic Programming (NeurIPS 2025 · similarity 0.85)
- Stochastic Zeroth-Order Optimization Under Heavy-Tailed Noise (NeurIPS 2026 · similarity 0.84)
- Double Variance Reduction: A Smoothing Trick for Composite Optimization Problems without First-Order Gradient (ICML 2024 · similarity 0.84)
- Non-Asymptotic and Non-Lipschitzian Bounds on Optimal Values in Stochastic Optimization Under Heavy Tails (ICML 2025 · similarity 0.84)
- Distributionally Robust Performative Optimization (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.
