# About Me

Zhewei Yao is a senior researcher and R&D manager at Microsoft, working on efficient large scale training and inference. He obtained his Ph.D. degree from University of California at Berkeley, where he was a Ph.D. researcher in BAIR, RISELab (former AMPLab), BDD, and Math Department. He was advised by Michael Mahoney, and he worked very closely with Kurt Keutzer. His research interest lies in computing statistics, optimization, and machine learning. Currently, he is interested in leveraging tools from randomized linear algebra to provide efficient and scalable solutions for large-scale optimization and learning problems. He is also working on the theory and application of deep learning. Before joining UC Berkeley, he received his B.S. in Math from Zhiyuan Honor College at Shanghai Jiao Tong University (last update 9/14/2022).

# Publications

## Conference

- ZeroQuant: Efficient and Affordable Post-Training Quantization for Large-Scale Transformers.

**Z. Yao**, R.Y. Aminabadi, M. Zhang, X. Wu, C. Li, Y. He

arXiv

Accepted for publication, Proc. NeurIPS 2022 - Extreme Compression for Pre-trained Transformers Made Simple and Efficient.

X. Wu^{*},**Z. Yao**, M. Zhang^{*}^{*}, C. Li, Y. He

arXiv

Accepted for publication, Proc. NeurIPS 2022 - DeepSpeed-MoE: Advancing Mixture-of-Experts Inference and Training to Power Next-Generation AI Scale.

S. Rajbhandari, C. Li,**Z. Yao**, M. Zhang, R. Y. Aminabadi, A. A. Awan, J. Rasley, Y. He

arXiv

Accepted for publication, Proc. ICML 2022 - Q-ASR: Integer-only Zero-shot Quantization for Efficient Speech Recognition.

S. Kim, A. Gholami,**Z. Yao**, A. Nrusimha, B. Zhai, T. Gao, M. W. Mahoney, K. Keutzer

arXiv

Accepted for publication, Proc. ICASSP 2022 - How Much Can CLIP Benefit Vision-and-Language Tasks?.

S. Shen, L. H. Li, H. Tan, M. Bansal, A. Rohrbach, K. Chang,**Z. Yao**, K Keutzer

arXiv

Accepted for publication, Proc. ICLR 2022 - Benchmarking Semi-supervised Federated Learning.

Z. Zhang^{*},**Z. Yao**, Y. Yang, Y. Yan, J. E. Gonzalez, and M. W. Mahoney^{*}

arXiv, code

Accepted for publication, Proc. IEEE BigData 2021 - Hessian-Aware Pruning and Optimal Neural Implant.

S. Yu^{*},**Z. Yao**, A. Gholami^{*}^{*}, Z. Dong^{*}, M. W. Mahoney, K. Keutzer

arXiv, code

Accepted for publication, Proc. WACV 2022 - What’s Hidden in a One-layer Randomly Weighted Transformer?.

S. Shen^{*},**Z. Yao**, D. Kiela, K. Keutzer, M. W. Mahoney^{*}

arXiv, code

Accepted for publication, Proc. EMNLP 2021 - ActNN: Reducing Training Memory Footprint via 2-Bit Activation Compressed Training.

J. Chen, L. Zheng, Z. Yao, D. Wang, I. Stoica, M. W. Mahoney, J. E. Gonzalez

arXiv

Accepted for publication, Proc. ICML 2021 (Long Talk). - I-BERT: Integer-only BERT Quantization.

S. Kim^{*}, A. Gholami^{*},**Z. Yao**, M. W. Mahoney, K. Keutzer^{*}

arXiv, code

Accepted for publication, Proc. ICML 2021 (Long Talk). - HAWQ-V3: Dyadic Neural Network Quantization.

**Z. Yao**, Z. Dong^{*}^{*}, Z. Zheng^{*}, A. Gholami^{*}, J. Yu, E. Tan, L. Wang, Q. Huang, Y. Wang, M. W. Mahoney, K. Keutzer

arXiv, code

Accepted for publication, Proc. ICML 2021 (Short Talk). - ADAHESSIAN: An Adaptive Second Order Optimizer for Machine Learning.

**Z. Yao**, A. Gholami^{*}^{*}, S. Shen, K. Keutzer, and M. W. Mahoney,

arXiv, code

Accepted for publication, Proc. AAAI 2021. - A Statistical Framework for Low-bitwidth Training of Deep Neural Networks.

J. Chen, Y. Gai,**Z. Yao**, M. W. Mahoney, and J. E. GonZalez

arXiv, code

Accepted for publication, Proc. NeurIPS 2020. - MAF: Multimodal Alignment Framework for Weakly-Supervised Phrase Grounding.

Q. Wang, H. Tan, S. Shen, M. W. Mahoney, and**Z. Yao**

arXiv, code

Accepted for publication, Proc. EMNLP 2020. - PowerNorm: Rethinking Batch Normalization in Transformers.

S. Shen^{*},**Z. Yao**, A. Gholami, M. W. Mahoney, and K. Keutzer^{*}

arXiv, code

Accepted for publication, Proc. ICML 2020. - ZeroQ: A Novel Zero Shot Quantization Framework.

Y. Cai^{*},**Z. Yao**, Z. Dong^{*}^{*}, A. Gholami, M. W. Mahoney, and K. Keutzer

arXiv, code

Accepted for publication, Proc. CVPR 2020. - PyHessian: Neural Networks Through the Lens of the Hessian.

**Z. Yao**, A. Gholami^{*}^{*}, K. Keutzer, M. W. Mahoney

arXiv, code

Accepted for publication, Proc. IEEE BigData 2020. - HAWQ-V2: Hessian Aware trace-Weighted Quantization of Neural Networks.

Z. Dong,**Z. Yao**, Y. Cai, D. Arfeen, A. Gholami, M. W. Mahoney, K. Keutzer

arXiv, code

Accepted for publication, Proc. NeurIPS 2020. - Q-BERT: Hessian Based Ultra Low Precision Quantization of BERT.

S. Shen, Z. Dong, J. Ye, L. Ma,**Z. Yao**, A. Gholami, M. W. Mahoney, K. Keutzer

arXiv

Accepted for publication, Proc. AAAI 2020. - ANODEV2: A Coupled Neural ODE Evolution Framework.

T. Zhang^{*},**Z. Yao**, A. Gholami^{*}^{*}, K. Keutzer, J. Gonzalez, G. Biros, and M. W. Mahoney

arXiv, code

Accepted for publication, Proc. NeurIPS 2019. - HAWQ: Hessian AWare Quantization of Neural Networks with Mixed-Precision.

Z. Dong^{*},**Z. Yao**, A. Gholami^{*}^{*}, M. W. Mahoney, K. Keutzer

arXiv, code

Accepted for publication, Proc. ICCV 2019. - Inefficiency of K-FAC for Large Batch Size Training.

L. Ma, G. Montague, J. Ye,**Z. Yao**, A. Gholami, K. Keutzer, M. W. Mahoney

arXiv

Accepted for publication, Proc. AAAI 2020. - JumpReLU: A Retrofit Defense Strategy for Adversarial Attacks.

N. B. Erichson^{*},**Z. Yao**, M. W. Mahoney^{*}

arXiv

Accepted for publication, Proc. ICPRAM 2020. - Trust Region Based Adversarial Attack on Neural Networks.

**Z. Yao**, A. Gholami, P. Xu, K. Keutzer, M. W. Mahoney

arXiv, code

Accepted for publication, Proc. CVPR 2019. - Hessian-based Analysis of Large Batch Training and Robustness to Adversaries.

**Z. Yao**, A. Gholami^{*}^{*}, Q. Lei, K. Keutzer, M. W. Mahoney

arXiv, code

Accepted for publication, Proc. NIPS 2018.

## Journal

- Inexact Newton-CG Algorithms With Complexity Guarantees.

**Z. Yao**, P. Xu, F. Roosta, S. J. Wright, M. W. Mahoney

arXiv

Accepted for publication, IMA Journal of Numerical Analysis (IMAJNA), 2022 - Shallow Learning for Fluid Flow Reconstruction with Limited Sensors and Limited Data.

N. B. Erichson, L. Mathelin,**Z. Yao**, S. L. Brunton, M. W. Mahoney, J. N. Kutz

arXiv

Accepted for publication, Proceedings of the Royal Society A. - Inexact non-convex Newton-type methods.

**Z. Yao**, P. Xu, F. Roosta-Khorasani, M. W. Mahoney

arXiv, code

Accepted for publication, INFORMS Journal on Optimization. - A hybrid adaptive MCMC algorithm in function spaces.

Q. Zhou, Z. Hu,**Z. Yao**, J. Li

arXiv

SIAM/ASA Journal on Uncertainty Quantification 5 (1), 621-639 - On an adaptive preconditioned Crank–Nicolson MCMC algorithm for infinite dimensional Bayesian inference.

Z. Hu,**Z. Yao**, J. Li

arXiv

Journal of Computational Physics 332, 492-503 - A TV-Gaussian prior for infinite-dimensional Bayesian inverse problems and its numerical implementation.

**Z. Yao**, Z. Hu, J. Li

arXiv

Inverse Problems 32 (7), 075006 (Highlight Paper)## Book Chapter

- A Survey of Quantization Methods for Efficient Neural Network Inference.

A. Gholami^{*}, S. Kim^{*}, Z. Dong^{*},**Z. Yao**, M. W. Mahoney, K. Keutzer^{*}

arXiv

Low-Power Computer Vision: Improving the Efficiency of Artificial Intelligence, 2021.

## Workshop

- Parameter Re-Initialization through Cyclical Batch Scheduling.

N. Mu^{*},**Z. Yao**, A. Gholami, K. Keutzer, M. W. Mahoney^{*}

arXiv

Accepted for publication, Proc. MLSYS Workshop at NIPS 2018 - An Empirical Exploration of Gradient Correlations in Deep Learning.

D. Rothchild, R. Fox, N. Golmant, J. Gonzalez, M. W. Mahoney, K. Rothauge, I. Stoica and**Z. Yao**

Integration of Deep Learning Theories, NeurIPS 2018

## Preprint and Technical Report

- BiFeat: Supercharge GNN Training via Graph Feature Quantization.

Y. Ma, P. Gong, J. Yi,**Z. Yao**, M. Wang, C. Li, Y. He, F. Yan

arXiv - MLPruning: A Multilevel Structured Pruning Framework for Transformer-based Models.

**Z. Yao**, L. Ma, S. Shen, K. Keutzer, M. W. Mahoney

arXiv - Residual Networks as Nonlinear Systems: Stability Analysis using Linearization.

K. Rothauge,**Z. Yao**, Z. Hu, and M. W. Mahoney

arXiv - On the Computational Inefficiency of Large Batch Sizes for Stochastic Gradient Descent.

N. Golmant, N. Vemuri,**Z. Yao**, V. Feinberg, A. Gholami, K. Rothauge, M. W. Mahoney, J. Gonzalez

arXiv - Large batch size training of neural networks with adversarial training and second-order information.

**Z. Yao**, A. Gholami^{*}^{*}, K. Keutzer, M. W. Mahoney

arXiv, code

# Selected Talks

- ICML’21 (ICML)

Online (Jul, 2021) - SIAM CSE’21: Beyond First Order Methods in Machine Learning Systems (CSE)

Online (Mar, 2021) - AAAI’21 (AAAI)

Online (Feb, 2021) - IEEE BigData’20 (BigData)

Online (Dec, 2020), slides - Berkeley Real-time Intelligent Secure Explanaible Systems Lab Camp (RiseLab)

Online (Oct, 2020), slides1 and slides2, vedio - Fast.AI (Fast.AI)

Online (Oct, 2020), slides, vedio - Scalable Parallel Computing Lab (SPCL)

Online (Oct, 2020), slides, vedio - ICML’20 Workshop on Beyond First-Order Optimization Methods in Machine Learning (Beyond)

Online (July, 2020), slides, vedio - Berkeley Real-time Intelligent Secure Explanaible Systems Lab Sponsor Retreat (RiseLab)

Tahoe Lake, CA, USA (May, 2020), slides - NeurIPS’19 Workshop on Beyond First-Order Optimization Methods in Machine Learning (Beyond)

Vancouver, Canada (December, 2019) - DIMACS Workshop on Randomized Numerical Linear Algebra, Statistics, and Optimization (DIMACS)

Rutgers University, New Jersey, USA (September, 2019), slides - Computer Vision Panel (IJCAI)

Macau, China (August, 2019), slides - Randomized Algorithms for Optimization Problems in Statistics (JSM)

Colorado Convention Center, Denver, Colorado, USA (July, 2019), slides - Berkeley Scientific Computing and Matrix Computations Seminar (Link)

Berkeley, CA, USA (November, 2018), slides - Berkeley Real-time Intelligent Secure Explanaible Systems Lab Sponsor Retreat (RiseLab)

Tahoe Lake, CA, USA (August, 2018), slides