Compress Deep Learning models 10,000x with Probabilistic Hash Functions - Livestream

Neural Scaling Law informally states that an increase in model size and data automatically improves AI. However, we have reached a point where the growth has reached a tipping end where the cost and energy associated with AI are becoming prohibitive.

This talk will demonstrate the algorithmic progress that can exponentially reduce the compute and memory cost of training and inference with neural networks. We will show how data structures,
particularly randomized hash tables, can be used to design an efficient "associative memory" that reduces the number of multiplications associated with the training of the neural networks.
Implementation of this algorithm challenges the common knowledge prevailing in the community that specialized processors like GPUs are significantly superior to CPUs for training large neural networks. The resulting algorithm is orders of magnitude cheaper and energy-efficient. Our careful implementations can train billions of parameter recommendations and NLP models on commodity desktop CPUs significantly faster than top-of-the-line TensorFlow alternatives on the most potent A100 GPU clusters, with the same or better accuracies. The same idea can also result in more than 50x faster and cheaper inference.

In the end, I will highlight a cache-friendly compression scheme that can compress embedding models by 10000x (100GB Embedding Table to 10MB) and still achieves the MLPerf benchmark AUC of 0.8025 on the Terabyte click-through Criteo data, getting significant inference speedup for free.

Speaker: Anshumali Shrivastava, Rice University

Register at weblink to receive connection information