Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Quasi-Self-Concordant Optimization with $\ell_{\infty}$ Lewis Weights
Authors: Alina Ene, Ta Duy Nguyen, Adrian Vladu
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We test our algorithm on ℓ2-regularized ℓp-regression problems: minx Ax b p p + µ Ax b 2 2. For p 3, the function |x|p + µx2 is C-QSC for C = pµ 1/(p 2). ... We compare our algorithm runtime against CVX and the Newton s method in [KSJ18]. ... On all small instances, our algorithm and the Newton s method have comparable runtime and are both significantly faster than CVX. |
| Researcher Affiliation | Academia | Alina Ene Department of Computer Science Boston University EMAIL Ta Duy Nguyen Department of Computer Science Boston University EMAIL Adrian Vladu CNRS & IRIF Université Paris Cité EMAIL |
| Pseudocode | Yes | Algorithm 1 ℓ -regression for ming x= 1 Ax Algorithm 2 ℓ -regression Subsolver(A, g, ε, M) Algorithm 3 Algorithm for optimizing h(x) = Pn i=1 f((Ax b)i) Algorithm 4 Residual Solver(x, M) Algorithm 5 Residual Solver(x, M) for n d Algorithm 6 Approximate ℓ Lewis weights Approx Lewis(A) |
| Open Source Code | Yes | Question: Does the paper provide open access to the data and code, with sufficient instructions to faithfully reproduce the main experimental results, as described in supplemental material? Answer: [Yes] Justification: We provide the code. |
| Open Datasets | Yes | Problems and datasets. We study the ℓp + ℓ2 regression problem in two settings: 1) Random matrix A, b: The entries of A, b are generated uniformly at random between 0 and 1, the second dimension (d) of A is fixed to 100; 2) On real world datasets: we use for real-world datasets available for regression tasks on UCI repository. The dataset sizes are reported in Table 1. Song Year Prediction [BM11], Consumption of Power [HB06], Protein Property [Ran13]. |
| Dataset Splits | No | Since CVX slows down significantly on larger instances, we only randomly pick up to 2500 in each instance. |
| Hardware Specification | Yes | Implementations were done on MATLAB 2024a on a Mac Book Pro M2/16GB RAM. |
| Software Dependencies | Yes | Implementations were done on MATLAB 2024a on a Mac Book Pro M2/16GB RAM. |
| Experiment Setup | Yes | In all experiments, we use p = 8 and precision ε = 10 10, µ = 1. |