PINCH: Predictive Importance-Sampling-Informed Cache for I/O-Bound DNN Training
DNN training on large datasets is often bottlenecked by data I/O rather than GPU computation. Shuffled SGD destroys temporal locality, rendering conventional caching and prefetching ineffective. Two classes of system address this: clairvoyant caches that exploit the deterministic access trace of seeded shuffling to pla...