Research map / Architectures and efficiency / RNNs and normalization
GPU-Accelerated Primal Learning for Extremely Fast Large-Scale Classification
John T Halloran, David M Rocke
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 |
|---|---|---|---|
| GPU-optimize TRON-LR by decoupling sequential dependencies and reducing transfers | TRON-LR · Sparse TRON-LR benchmarks | average training time improvement over single-threaded TRON-LR: 70.8% | Improved |
| combine GPU optimization with multithreading | TRON-LR · Sparse TRON-LR benchmarks | average training time improvement over single-thread-optimized TRON: 89.2% | Improved |
| combine GPU optimization with multithreading | TRON-LR · Sparse TRON-LR benchmarks | average training time improvement over multithread-optimized TRON: 65.2% | Improved |
| use the direct GPU implementation with the original variable access pattern (ablation) | TRON-LR · SUSY | speedup: 0.65 | Hurt |
| use the mixed GPU and multithread architecture | TRON-LR · Sparse TRON-LR benchmarks | average speedup: 9.3 fold | Improved |
| use the GPU-optimized SVM solver | Percolator SVM learning · Kim | speedup over original Percolator SVM learning engine: 7.4 fold | Improved |
| use the mixed GPU and multithread SVM solver | Percolator SVM learning · Kim | speedup over original Percolator SVM learning engine: 6.6 fold | Improved |
| use the mixed GPU and multithread SVM solver | Percolator SVM learning · Wilhelm | speedup over original Percolator SVM learning engine: 5.4 fold | Improved |
| move dense SVM computations to GPU while using multithreading for memory-intensive active-set operations | Percolator SVM learning · Wilhelm | SVM learning time: 4.4 days -> 19.7 hours | Improved |
| use PyTorch GPU gradient descent rather than CPU gradient descent | PyTorch logistic regression · rcv1 | GPU speedup: 14.63 | Improved |
| use PyTorch GPU L-BFGS rather than CPU L-BFGS | PyTorch logistic regression · rcv1 | GPU speedup: 13.08 | Improved |
Closest papers by meaning
- Parallelizing non-linear sequential models over the sequence length (ICLR 2024 · similarity 0.83)
- Navigating Extremes: Dynamic Sparsity in Large Output Spaces (NeurIPS 2024 · similarity 0.83)
- KernelBench: Can LLMs Write Efficient GPU Kernels? (ICML 2025 · similarity 0.83)
- Pfeife: Automatic Pipeline Parallelism for PyTorch (ICML 2025 · similarity 0.83)
- 70% Size, 100% Accuracy: Lossless LLM Compression for Efficient GPU Inference via Dynamic-Length Float (DFloat11) (NeurIPS 2025 · similarity 0.82)
- Harnessing Manycore Processors with Distributed Memory for Accelerated Training of Sparse and Recurrent Models (AAAI 2024 · 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.
