AIBullisharXiv – CS AI · 7h ago7/10
🧠
HASTE: Hardware-Aware Dynamic Sparse Training for Large Output Spaces
Researchers introduce HASTE, a hardware-aware sparse training method for extreme multi-label classification that uses group-shared fixed fan-in sparsity to optimize GPU execution. The approach achieves up to 25x speedup in backward passes compared to standard sparse methods while maintaining competitive accuracy, addressing the memory-compute bottleneck in models with millions of output labels.