Answers are grounded in this paper and cite their sources.
Most scientific and AI applications rely on data that is usually sparse. Sparsity is sometimes deliberately introduced with pruning of insignificant and redundant connections in the neural network layers. This sparse data can efficiently be represented through various compression formats, thereby, reducing storage footprint and eliminating computations with zero values. However, accessing nonzero values introduces extra computational overhead.
No takes yet. Share an insight, caveat, or question.
Vasireddy et al. (2023) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: