Get Started
Research questionHow can self-organizing maps scale to millions of neurons within GPU memory and bandwidth limits?At MEDLINE scale, each training epoch repeatedly searches a large codebook for every sample, making memory traffic and codebook storage major costs. These constraints limit both training speed and the number of neurons that can fit on one GPU.
Machine Learning
Neural and Evolutionary Computing
Latest papersRecent research connected to this question, newest first.A Feature-Major Codebook for Memory-Efficient Sparse-Binary Self-Organizing Maps: Scaling a MEDLINE Atlas to 1.05 Million Neurons on a Single Consumer GPUThe evidence concerns sparse-binary self-organizing maps trained on 29.9 million MEDLINE articles, with benchmarks on a 24 GB consumer GPU and a 141 GB H200. It reports runtime, memory feasibility, and held-out quantisation error for maps up to 1,048,576 neurons, compared with specified sparse-GPU and CPU implementations; the addendum changes some comparative speed results and narrows two mechanism claims.research paper · Sep 2, 2026
Related questions
How should GPU implementations be compared fairly when their optimization effort is asymmetric?Can text-to-image models match web-scale performance using smaller, reproducible datasets and models?How can graph neural networks capture long-range interactions without oversmoothing or oversquashing on diverse large graphs?How can neural networks retain capacity while fitting within fixed parameter and memory budgets?
Home
Topics
Search
Library