Research map / Label-efficient and robust learning / Self-supervised and contrastive
torchsom: The Reference PyTorch Library for Self-Organizing Maps
Louis Berthier, Ahmed Shokry, Maxime Moreaud, Guillaume Ramelet, Eric Moulines
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 torchsom instead of standard MiniSom | SOM training · 25x15 rectangular, 16000 x 300, CPU | topographic error (%): 76 -> 6 | Improved |
| use torchsom instead of standard MiniSom | SOM training · 25x15 rectangular, 16000 x 300, CPU | wall-clock time (s): 900 -> 22.68 | Improved |
| use torchsom instead of MiniSom-JIT | SOM training · 25x15 rectangular, 16000 x 300, CPU | wall-clock time (s): 276 -> 22.68 | Improved |
| use torchsom instead of MiniSom | SOM training · 25x15 rectangular, CPU | quantization error: 0.17 -> 0.24 | Mixed |
| use torchsom GPU instead of somoclu GPU on small low-dimensional data | SOM training · 25x15 rectangular, 240 x 4, GPU | wall-clock time (s): 0.15 -> 1.29 | Hurt |
| use torchsom GPU instead of somoclu GPU on large high-dimensional data | SOM training · 25x15 rectangular, 16000 x 300, GPU | wall-clock time (s): 161 -> 13.08 | Improved |
| use torchsom GPU instead of somoclu GPU | SOM training · 25x15 rectangular, 16000 x 300, GPU | topographic error (%): 37 -> 5 | Improved |
| scale map from 25x15 to 90x70 for torchsom vs MiniSom | SOM training · 90x70 rectangular, 16000 x 300, CPU | wall-clock time (s): 25860 -> 130 | Improved |
| scale map to 90x70 for torchsom GPU vs somoclu | SOM training · 90x70 rectangular, 16000 x 300, GPU | wall-clock time (s): 1038 -> 15.64 | Improved |
| use hexagonal instead of rectangular topology in torchsom | torchsom GPU · 16000 x 300, 25x15 map | wall-clock time (s): 15.6 -> 94.8 | Hurt |
| use torchsom GPU instead of CPU | torchsom · 25x15 rectangular, 16000 x 300 | wall-clock time (s): 22.68 -> 13.08 | Improved |
| use torchsom instead of MiniSom-JIT | SOM training · 25x15 hexagonal, 240 x 4, CPU | topographic error (%): 22 -> 32 | Hurt |
Closest papers by meaning
- Torchhd: An Open Source Python Library to Support Research on Hyperdimensional Computing and Vector Symbolic Architectures (JMLR 2023 · similarity 0.84)
- torchgfn: A PyTorch GFlowNet Library (JMLR 2026 · similarity 0.83)
- TorchSpatial: A Location Encoding Framework and Benchmark for Spatial Representation Learning (NeurIPS 2024 · similarity 0.81)
- TorchRL: A data-driven decision-making library for PyTorch (ICLR 2024 · similarity 0.80)
- The Map Equation Goes Neural: Mapping Network Flows with Graph Neural Networks (NeurIPS 2024 · similarity 0.80)
- Pfeife: Automatic Pipeline Parallelism for PyTorch (ICML 2025 · similarity 0.80)
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.
