AI Paper Atlas

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

JMLR 2026 · Label-efficient and robust learning › Self-supervised and contrastive · approach: Invariance-based self-supervised learning

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 changedWhereMeasuredEffect
use torchsom instead of standard MiniSomSOM training · 25x15 rectangular, 16000 x 300, CPUtopographic error (%): 76 -> 6Improved
use torchsom instead of standard MiniSomSOM training · 25x15 rectangular, 16000 x 300, CPUwall-clock time (s): 900 -> 22.68Improved
use torchsom instead of MiniSom-JITSOM training · 25x15 rectangular, 16000 x 300, CPUwall-clock time (s): 276 -> 22.68Improved
use torchsom instead of MiniSomSOM training · 25x15 rectangular, CPUquantization error: 0.17 -> 0.24Mixed
use torchsom GPU instead of somoclu GPU on small low-dimensional dataSOM training · 25x15 rectangular, 240 x 4, GPUwall-clock time (s): 0.15 -> 1.29Hurt
use torchsom GPU instead of somoclu GPU on large high-dimensional dataSOM training · 25x15 rectangular, 16000 x 300, GPUwall-clock time (s): 161 -> 13.08Improved
use torchsom GPU instead of somoclu GPUSOM training · 25x15 rectangular, 16000 x 300, GPUtopographic error (%): 37 -> 5Improved
scale map from 25x15 to 90x70 for torchsom vs MiniSomSOM training · 90x70 rectangular, 16000 x 300, CPUwall-clock time (s): 25860 -> 130Improved
scale map to 90x70 for torchsom GPU vs somocluSOM training · 90x70 rectangular, 16000 x 300, GPUwall-clock time (s): 1038 -> 15.64Improved
use hexagonal instead of rectangular topology in torchsomtorchsom GPU · 16000 x 300, 25x15 mapwall-clock time (s): 15.6 -> 94.8Hurt
use torchsom GPU instead of CPUtorchsom · 25x15 rectangular, 16000 x 300wall-clock time (s): 22.68 -> 13.08Improved
use torchsom instead of MiniSom-JITSOM training · 25x15 hexagonal, 240 x 4, CPUtopographic error (%): 22 -> 32Hurt

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy