Beidi Chen's guest appearances

Beidi Chen

The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence)

Stanford MLSys Seminar

Virginia Smith

Alex Ratner

Scott Clark

Celeste Kidd

Dennis Mortensen

Rob Walker

Sanmi Koyejo

Chelsea Finn

Dario Amodei

Jianxiong Xiao

Kate Darling

Jason Holmberg

Lotte Bransen

Christopher Berner

David Spiegelhalter

OwlTail

Beidi Chen talks about "Pixelated Butterfly: Simple and Efficient Sparse Training for Neural Network Models." Overparameterized neural networks generalize well but are expensive to train. Ideally, one would like to reduce their computational cost while retaining their generalization benefits. Sparse model training is a simple and promising approach to achieve this, but there remain challenges as existing methods struggle with accuracy loss, slow training runtime, or difficulty in sparsifying all model components. The core problem is that searching for a sparsity mask over a discrete set of sparse matrices is difficult and expensive. To address this, our main insight is to optimize over a continuous superset of sparse matrices with a fixed structure known as products of butterfly matrices. As butterfly matrices are not hardware efficient, we propose simple variants of butterfly (block and flat) to take advantage of modern hardware. Our method (Pixelated Butterfly) uses a simple fixed sparsity pattern based on flat block butterfly and low-rank matrices to sparsify most network layers (e.g., attention, MLP). We empirically validate that Pixelated Butterfly is 3x faster than butterfly and speeds up training to achieve favorable accuracy--efficiency tradeoffs. On the ImageNet classification and WikiText-103 language modeling tasks, our sparse models train up to 2.5x faster than the dense MLP-Mixer, Vision Transformer, and GPT-2 medium with no drop in accuracy.

<p>Beidi Chen talks about "Pixelated Butterfly: Simple and Efficient Sparse Training for Neural Network Models." Overparameterized neural networks generalize well but are expensive to train. Ideally, one would like to reduce their computational cost while retaining their generalization benefits. Sparse model training is a simple and promising approach to achieve this, but there remain challenges as existing methods struggle with accuracy loss, slow training runtime, or difficulty in sparsifying all model components. The core problem is that searching for a sparsity mask over a discrete set of sparse matrices is difficult and expensive. To address this, our main insight is to optimize over a continuous superset of sparse matrices with a fixed structure known as products of butterfly matrices. As butterfly matrices are not hardware efficient, we propose simple variants of butterfly (block and flat) to take advantage of modern hardware. Our method (Pixelated Butterfly) uses a simple fixed sparsity pattern based on flat block butterfly and low-rank matrices to sparsify most network layers (e.g., attention, MLP). We empirically validate that Pixelated Butterfly is 3x faster than butterfly and speeds up training to achieve favorable accuracy--efficiency tradeoffs. On the ImageNet classification and WikiText-103 language modeling tasks, our sparse models train up to 2.5x faster than the dense MLP-Mixer, Vision Transformer, and GPT-2 medium with no drop in accuracy.</p>

01/06/22 #49 Beidi Chen - Pixelated Butterfly: Fast Machine Learning with Sparsity

Beidi Chen is part of the team that developed a cheaper, algorithmic, CPU alternative to state-of-the-art GPU machines. They presented their findings at NeurIPS 2019 and have since gained a lot of attention for their paper, SLIDE: In Defense of Smart Algorithms Over Hardware Acceleration for Large-Scale Deep Learning Systems. Beidi shares how the team took a new look at deep learning with the case of extreme classification by turning it into a search problem and using locality-sensitive hashing.

<p>Beidi Chen is part of the team that developed a cheaper, algorithmic, CPU alternative to state-of-the-art GPU machines. They presented their findings at NeurIPS 2019 and have since gained a lot of attention for their paper, SLIDE: In Defense of Smart Algorithms Over Hardware Acceleration for Large-Scale Deep Learning Systems. Beidi shares how the team took a new look at deep learning with the case of extreme classification by turning it into a search problem and using locality-sensitive hashing.</p>

SLIDE: Smart Algorithms over Hardware Acceleration for Large-Scale Deep Learning with Beidi Chen - #356