Research map / Learning theory and algorithms / Combinatorial and search optimization
Optimization for Classical Machine Learning Problems on the GPU
Sören Laue, Mark Blacher, Joachim Giesen
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 the GPU-efficient GENO solver instead of cuOSQP | Dual SVM · cod-rna | running time (seconds): 55.7 -> 0.6 | Improved |
| Use the GPU-efficient GENO solver instead of SCS GPU | Dual SVM · cod-rna | running time (seconds): 2342.0 -> 0.6 | Improved |
| Use the GPU-efficient GENO solver instead of Fairlearn | Fairness optimization · adult, logistic constraint, 42500 samples | running time (seconds): $30.0\pm 0.2$ -> $0.8\pm 0.0$ | Improved |
| Use the GPU-efficient GENO solver instead of cuOSQP | Dual SVM · adult a1a–a8a, 11220 samples | running time (seconds): $42.9$ -> $0.4$ | Improved |
| Run the GENO solver on GPU rather than CPU | NNLS · data set 1, 10000 | running time (seconds): $3.7\pm 0.4$ -> $0.8\pm 0.1$ | Improved |
| Use the GPU-efficient GENO solver instead of GENO CPU | NNLS · data set 2, 44400 | running time (seconds): $6.4\pm 0.1$ -> $1.6\pm 0.2$ | Improved |
| Run the GENO solver on GPU rather than CPU | Joint probability optimization · entropy regularizer, data set 1, 1000 | running time (seconds): $4.6$ -> $1.1$ | Improved |
| Use the GPU-efficient GENO solver instead of GENO CPU | Joint probability optimization · Gaussian regularizer, data set 2, 3000 | running time (seconds): $2622.4\pm 236.7$ -> $92.3\pm 1.9$ | Improved |
| Replace original L-BFGS-B with the GPU-efficient L-BFGS-B method, avoiding sequential Cauchy point computation | L-BFGS-B on GPU · NNLS, problem size 12,000 | total running time (seconds): 5.2 -> 0.8 | Improved |
| Port original L-BFGS-B from CPU to GPU, retaining its sequential Cauchy point computation | Original L-BFGS-B · NNLS, problem size 12,000 | total running time (seconds): 4.9 -> 5.2 | Hurt |
Closest papers by meaning
- GENO -- GENeric Optimization for Classical Machine Learning (NeurIPS 2019 · similarity 0.87)
- A GPU-based Constraint Programming Solver (AAAI 2026 · similarity 0.86)
- DOGE-Train: Discrete Optimization on GPU with End-to-End Training (AAAI 2024 · similarity 0.84)
- B$^3$-PWL: GPU-Batched Branch-and-Bound for Piecewise-Linear Optimization with SOS2 Constraints (NeurIPS 2026 · similarity 0.83)
- StitchCUDA: An Automated Multi-Agents End-to-End GPU Programing Framework with Rubric-based Agentic Reinforcement Learning (ICML 2026 · similarity 0.82)
- FlashRNN: I/O-Aware Optimization of Traditional RNNs on modern hardware (ICLR 2025 · similarity 0.82)
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.
