Alert button
Picture for Michael W. Mahoney

Michael W. Mahoney

Alert button

Taxonomizing local versus global structure in neural network loss landscapes

Jul 23, 2021
Yaoqing Yang, Liam Hodgkinson, Ryan Theisen, Joe Zou, Joseph E. Gonzalez, Kannan Ramchandran, Michael W. Mahoney

Figure 1 for Taxonomizing local versus global structure in neural network loss landscapes
Figure 2 for Taxonomizing local versus global structure in neural network loss landscapes
Figure 3 for Taxonomizing local versus global structure in neural network loss landscapes
Figure 4 for Taxonomizing local versus global structure in neural network loss landscapes
Viaarxiv icon

Newton-LESS: Sparsification without Trade-offs for the Sketched Newton Update

Jul 15, 2021
Michał Dereziński, Jonathan Lacotte, Mert Pilanci, Michael W. Mahoney

Figure 1 for Newton-LESS: Sparsification without Trade-offs for the Sketched Newton Update
Figure 2 for Newton-LESS: Sparsification without Trade-offs for the Sketched Newton Update
Figure 3 for Newton-LESS: Sparsification without Trade-offs for the Sketched Newton Update
Figure 4 for Newton-LESS: Sparsification without Trade-offs for the Sketched Newton Update
Viaarxiv icon

Compressing Deep ODE-Nets using Basis Function Expansions

Jun 21, 2021
Alejandro Queiruga, N. Benjamin Erichson, Liam Hodgkinson, Michael W. Mahoney

Figure 1 for Compressing Deep ODE-Nets using Basis Function Expansions
Figure 2 for Compressing Deep ODE-Nets using Basis Function Expansions
Figure 3 for Compressing Deep ODE-Nets using Basis Function Expansions
Figure 4 for Compressing Deep ODE-Nets using Basis Function Expansions
Viaarxiv icon

Post-mortem on a deep learning contest: a Simpson's paradox and the complementary roles of scale metrics versus shape metrics

Jun 01, 2021
Charles H. Martin, Michael W. Mahoney

Figure 1 for Post-mortem on a deep learning contest: a Simpson's paradox and the complementary roles of scale metrics versus shape metrics
Figure 2 for Post-mortem on a deep learning contest: a Simpson's paradox and the complementary roles of scale metrics versus shape metrics
Figure 3 for Post-mortem on a deep learning contest: a Simpson's paradox and the complementary roles of scale metrics versus shape metrics
Figure 4 for Post-mortem on a deep learning contest: a Simpson's paradox and the complementary roles of scale metrics versus shape metrics
Viaarxiv icon

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

May 30, 2021
Zhewei Yao, Linjian Ma, Sheng Shen, Kurt Keutzer, Michael W. Mahoney

Figure 1 for MLPruning: A Multilevel Structured Pruning Framework for Transformer-based Models
Figure 2 for MLPruning: A Multilevel Structured Pruning Framework for Transformer-based Models
Figure 3 for MLPruning: A Multilevel Structured Pruning Framework for Transformer-based Models
Figure 4 for MLPruning: A Multilevel Structured Pruning Framework for Transformer-based Models
Viaarxiv icon

ActNN: Reducing Training Memory Footprint via 2-Bit Activation Compressed Training

Apr 29, 2021
Jianfei Chen, Lianmin Zheng, Zhewei Yao, Dequan Wang, Ion Stoica, Michael W. Mahoney, Joseph E. Gonzalez

Figure 1 for ActNN: Reducing Training Memory Footprint via 2-Bit Activation Compressed Training
Figure 2 for ActNN: Reducing Training Memory Footprint via 2-Bit Activation Compressed Training
Figure 3 for ActNN: Reducing Training Memory Footprint via 2-Bit Activation Compressed Training
Figure 4 for ActNN: Reducing Training Memory Footprint via 2-Bit Activation Compressed Training
Viaarxiv icon

Q-ASR: Integer-only Zero-shot Quantization for Efficient Speech Recognition

Mar 31, 2021
Sehoon Kim, Amir Gholami, Zhewei Yao, Anirudda Nrusimha, Bohan Zhai, Tianren Gao, Michael W. Mahoney, Kurt Keutzer

Figure 1 for Q-ASR: Integer-only Zero-shot Quantization for Efficient Speech Recognition
Figure 2 for Q-ASR: Integer-only Zero-shot Quantization for Efficient Speech Recognition
Figure 3 for Q-ASR: Integer-only Zero-shot Quantization for Efficient Speech Recognition
Figure 4 for Q-ASR: Integer-only Zero-shot Quantization for Efficient Speech Recognition
Viaarxiv icon

A Survey of Quantization Methods for Efficient Neural Network Inference

Mar 25, 2021
Amir Gholami, Sehoon Kim, Zhen Dong, Zhewei Yao, Michael W. Mahoney, Kurt Keutzer

Figure 1 for A Survey of Quantization Methods for Efficient Neural Network Inference
Figure 2 for A Survey of Quantization Methods for Efficient Neural Network Inference
Figure 3 for A Survey of Quantization Methods for Efficient Neural Network Inference
Figure 4 for A Survey of Quantization Methods for Efficient Neural Network Inference
Viaarxiv icon

Hessian Eigenspectra of More Realistic Nonlinear Models

Mar 17, 2021
Zhenyu Liao, Michael W. Mahoney

Figure 1 for Hessian Eigenspectra of More Realistic Nonlinear Models
Figure 2 for Hessian Eigenspectra of More Realistic Nonlinear Models
Figure 3 for Hessian Eigenspectra of More Realistic Nonlinear Models
Figure 4 for Hessian Eigenspectra of More Realistic Nonlinear Models
Viaarxiv icon