Timezone: »

Accelerated Sparse Neural Training: A Provable and Efficient Method to Find N:M Transposable Masks
Itay Hubara · Brian Chmiel · Moshe Island · Ron Banner · Joseph Naor · Daniel Soudry

Thu Dec 09 08:30 AM -- 10:00 AM (PST) @

Unstructured pruning reduces the memory footprint in deep neural networks (DNNs). Recently, researchers proposed different types of structural pruning intending to reduce also the computation complexity. In this work, we first suggest a new measure called mask-diversity which correlates with the expected accuracy of the different types of structural pruning. We focus on the recently suggested N:M fine-grained block sparsity mask, in which for each block of M weights, we have at least N zeros. While N:M fine-grained block sparsity allows acceleration in actual modern hardware, it can be used only to accelerate the inference phase. In order to allow for similar accelerations in the training phase, we suggest a novel transposable fine-grained sparsity mask, where the same mask can be used for both forward and backward passes. Our transposable mask guarantees that both the weight matrix and its transpose follow the same sparsity pattern; thus, the matrix multiplication required for passing the error backward can also be accelerated. We formulate the problem of finding the optimal transposable-mask as a minimum-cost flow problem. Additionally, to speed up the minimum-cost flow computation, we also introduce a fast linear-time approximation that can be used when the masks dynamically change during training. Our experiments suggest a 2x speed-up in the matrix multiplications with no accuracy degradation over vision and language models. Finally, to solve the problem of switching between different structure constraints, we suggest a method to convert a pre-trained model with unstructured sparsity to an N:M fine-grained block sparsity model with little to no training. A reference implementation can be found at https://github.com/papers-submission/structuredtransposablemasks.

Author Information

Itay Hubara (Habana Labs)
Brian Chmiel (Intel)
Moshe Island
Ron Banner (Intel - Artificial Intelligence Products Group (AIPG))
Joseph Naor (Technion, Technion)
Daniel Soudry (Technion)

I am an assistant professor in the Department of Electrical Engineering at the Technion, working in the areas of Machine learning and theoretical neuroscience. I am especially interested in all aspects of neural networks and deep learning. I did my post-doc (as a Gruss Lipper fellow) working with Prof. Liam Paninski in the Department of Statistics, the Center for Theoretical Neuroscience the Grossman Center for Statistics of the Mind, the Kavli Institute for Brain Science, and the NeuroTechnology Center at Columbia University. I did my Ph.D. (2008-2013, direct track) in the Network Biology Research Laboratory in the Department of Electrical Engineering at the Technion, Israel Institute of technology, under the guidance of Prof. Ron Meir. In 2008 I graduated summa cum laude with a B.Sc. in Electrical Engineering and a B.Sc. in Physics, after studying in the Technion since 2004.

More from the Same Authors