Research map / Architectures and efficiency / RNNs and normalization
Kernel Operations on the GPU, with Autodiff, without Memory Overflows
Benjamin Charlier, Jean Feydy, Joan Alexis Glaunès, François-David Collin, Ghislain Durif
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 PyKeOps symbolic reduction instead of PyTorch dense tensors | Gaussian kernel matvec · N=10k, D=3 | runtime (ms): 9 -> 0.7 | Improved |
| use PyKeOps instead of PyTorch dense tensors at larger size | Gaussian kernel matvec · N=100k, D=3 | memory: out of mem -> 15.0 | Improved |
| use PyKeOps instead of TF-XLA | Gaussian kernel matvec · N=100k, D=3 | runtime (ms): 89 -> 15.0 | Improved |
| use PyKeOps instead of Halide | Gaussian kernel matvec · N=1M, D=3 | runtime (s): 3.8 -> 1.39 | Improved |
| use PyKeOps instead of TVM | Gaussian kernel matvec · N=1M, D=3 | runtime (s): 2.79 -> 1.39 | Improved |
| use hand-written C++ KeOps++ instead of PyKeOps | Gaussian kernel matvec · N=10k, D=3 | runtime (ms): 0.7 -> 0.4 | Improved |
| use hand-written C++ KeOps++ instead of PyKeOps | Gaussian kernel matvec · N=1M, D=3 | runtime (s): 1.39 -> 1.38 | No clear effect |
| use PyKeOps instead of Halide for code length | Gaussian kernel matvec · code size | lines of code: 15 -> 5 | Improved |
| use PyKeOps instead of TVM for code length | Gaussian kernel matvec · code size | lines of code: 17 -> 5 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- PyTorch — PyTorch: An Imperative Style, High-Performance Deep Learning Library (NeurIPS 2019) · better on 1 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Giga-scale Kernel Matrix-Vector Multiplication on GPU (NeurIPS 2022 · similarity 0.82)
- Kernel Methods Through the Roof: Handling Billions of Points Efficiently (NeurIPS 2020 · similarity 0.81)
- Fast Inference with Kronecker-Sparse Matrices (ICML 2025 · similarity 0.80)
- Einsum Benchmark: Enabling the Development of Next-Generation Tensor Execution Engines (NeurIPS 2024 · similarity 0.80)
- KernelBench: Can LLMs Write Efficient GPU Kernels? (ICML 2025 · similarity 0.79)
- Column-Oriented Datalog on the GPU (AAAI 2025 · similarity 0.79)
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.
