Search papers, labs, and topics across Lattice.
This paper introduces a feature-major codebook layout for self-organizing maps (SOMs) that significantly enhances the efficiency of best-matching-unit (BMU) searches, enabling the construction of a MEDLINE atlas with over a million neurons on a single consumer GPU. By recasting the BMU search as a tiled sparse-dense product, the proposed method achieves speedups of 4.5-8.5x compared to existing implementations without sacrificing quantization accuracy. The resulting SparseBin.SOM not only accommodates larger map sizes but also demonstrates remarkable efficiency, training a converged map of 29.9 million MEDLINE articles in just 72 seconds on a 24 GB GPU.
A new feature-major codebook layout accelerates self-organizing map training by up to 621x, enabling the largest reported atlas of 1.05 million neurons on a single GPU.
A self-organising map turns a large corpus into a browsable two-dimensional atlas, but building one at MEDLINE scale has been impractical: the best-matching-unit (BMU) search that dominates training is bound by the bandwidth needed to read the codebook every epoch. I show that this bottleneck is largely an artefact of codebook layout. Storing it feature-major with each feature's weights contiguous, W[v.M+i], recasts the search as a tiled sparse-dense product in which every loaded weight column is reused across a tile of samples. Varying only the layout, with implementation, precision and update rule held fixed, accelerates the BMU search by 4.5-8.5x. Because an exact-argmin BMU is invariant to how the codebook is stored, this gain costs nothing: held-out quantisation error agrees with a cuSPARSE baseline to within 0.5% at every map size. Against that baseline the advantage is a crossover rather than a constant: cuSPARSE.SOM is faster at small maps, SparseBin.SOM is 1.5x faster at 128x128 and 2.6x at 256x256, and at 512x512 it is the only one that runs at all on 24 GB. Paired with a radius-independent box-blur update and a convergence-based stopping rule, it trains a converged map over 29.9 million MEDLINE articles in about 72 s at 64x64 on one 24 GB GPU, and accommodates 262,144 neurons (512x512 edges) where every alternative algorithm I tested exceeds memory constraints. On a 141 GB H200 it reaches 1,048,576 neurons (1024x1024 edges) - to my knowledge the largest self-organising map yet reported. Held-out error follows a smooth power law with no elbow across three decades of map size, so the limit on resolution is compute rather than any breakpoint in the data. At matched work the design is ~82x faster than MedSOM, the CUDA implementation behind our earlier MEDLINE atlases and, at 128x128, 621x faster than the best available multicore-CPU library.
