Pith. sign in

REVIEW 1 cited by

Towards Memory-Efficient Training for Extremely Large Output Spaces -- Learning with 500k Labels on a Single Commodity GPU

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2306.03725 v1 pith:WQYQ7PYA submitted 2023-06-06 cs.LG cs.AIcs.DC

classification cs.LGcs.AIcs.DC
keywords labelslayermemoryoutputsparsecommodityconnectivitylarge
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
read the original abstract

In classification problems with large output spaces (up to millions of labels), the last layer can require an enormous amount of memory. Using sparse connectivity would drastically reduce the memory requirements, but as we show below, it can result in much diminished predictive performance of the model. Fortunately, we found that this can be mitigated by introducing a penultimate layer of intermediate size. We further demonstrate that one can constrain the connectivity of the sparse layer to be uniform, in the sense that each output neuron will have the exact same number of incoming connections. This allows for efficient implementations of sparse matrix multiplication and connection redistribution on GPU hardware. Via a custom CUDA implementation, we show that the proposed approach can scale to datasets with 670,000 labels on a single commodity GPU with only 4GB memory.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. NeuroTrails: Training with Dynamic Sparse Heads as the Key to Effective Ensembling

    cs.LG 2025-05 conditional novelty 6.0 of 10

    Dynamic sparse training of multiple heads on a shared backbone outperforms full dense ensembles on ImageNet and C4 while using less compute.

Pith tools