Heterogeneous Quantization

Heterogeneous quantization optimizes deep neural network efficiency by assigning different bit-precisions to various model parameters and operations, aiming to minimize accuracy loss while maximizing speed and memory savings. Current research focuses on developing automated methods for determining optimal bit-width assignments across diverse model architectures, including transformers and convolutional neural networks, and addressing quantization-induced inefficiencies in specific layers like batch normalization. This approach is crucial for deploying deep learning models on resource-constrained edge devices, enabling broader applications in areas like embedded systems and mobile computing.

Papers

October 29, 2024

Data Generation for Hardware-Friendly Post-Training Quantization
Lior Dikstein, Ariel Lapid, Arnon Netzer, Hai Victor Habi
Quantization Operator Post Training Quantization Data Generation Zero Shot Quantization Quantization Performance Heterogeneous Quantization

October 4, 2024

Resource-aware Mixed-precision Quantization for Enhancing Deployability of Transformers for Time-series Forecasting on Embedded FPGAs
Tianheng Ling, Chao Qian, Gregor Schiele
Transformer Megatron Decepticons Time Series Forecasting Field Programmable Gate Array Mixed Precision Quantization Uncertainty Aware Deployment FPGA Resource Heterogeneous Quantization

September 27, 2024

Heterogeneous quantization regularizes spiking neural network activity
Roy Moyal, Kyrus R. Mama, Matthew Einhorn, Ayon Borthakur, Thomas A. Cleland
Internal Representation Neural Activity Synaptic Weight Heterogeneous Quantization Stimulus Optimization

March 29, 2024

PikeLPN: Mitigating Overlooked Inefficiencies of Low-Precision Neural Networks
Marina Neseem, Conor McCullough, Randy Hsin, Chas Leichner, Shan Li, In Suk Chong, Andrew G. Howard, Lukasz Lew, Sherief Reda, Ville-Mikko Rautio, Daniele Moro
High Efficiency Batch Normalization Low Precision Batch Normalization Layer Multiply Accumulate Low Precision Quantization Heterogeneous Quantization

November 10, 2023

Automated Heterogeneous Low-Bit Quantization of Multi-Model Deep Learning Inference Pipeline
Jayeeta Mondal, Swarnava Dey, Arijit Mukherjee
Bit Quantization Multiple Deep Neural Network Single Deep Inference Pipeline Heterogeneous Quantization

September 5, 2023

On-Chip Hardware-Aware Quantization for Mixed Precision Neural Networks
Wei Huang, Haotong Qin, Yangdong Liu, Jingzhuo Liang, Yulun Zhang, Ying Li, Xianglong Liu
Mixed Precision Quantization Mixed Precision Heterogeneous Quantization Quantization Aware Better on Chip Data

February 24, 2023

DyBit: Dynamic Bit-Precision Numbers for Efficient Quantized Neural Network Inference
Jiajun Zhou, Jiajun Wu, Yizhao Gao, Yuhao Ding, Chaofan Tao, Boyu Li, Fengbin Tu, Kwang-Ting Cheng, Hayden Kwok-Hay So, Ngai Wong
DNN Model Neural Network Weight Low Bit Heterogeneous Quantization

December 3, 2022

Make RepVGG Greater Again: A Quantization-aware Approach
Xiangxiang Chu, Liang Li, Bo Zhang
Supervised ImageNet Post Training Quantization Quantization Performance Heterogeneous Quantization

October 7, 2022

A Closer Look at Hardware-Friendly Weight Quantization
Sungmin Bae, Piotr Zielinski, Satrajit Chatterjee
Glance Annotation Quantization Error Weight Quantization Heterogeneous Quantization

June 15, 2022

Edge Inference with Fully Differentiable Quantized Mixed Precision Neural Networks
Clemens JS Schaefer, Siddharth Joshi, Shan Li, Raul Blazquez
Quantization Technique Quantization Error Closed Form Differentiable Expression Mixed Precision Edge Inference Quantization Learning Heterogeneous Quantization

May 17, 2022

A Silicon Photonic Accelerator for Convolutional Neural Networks with Heterogeneous Quantization
Febin Sunny, Mahdi Nikdast, Sudeep Pasricha
Convolutional Neural Network Photonic Accelerator CNN Accelerator Silicon Photonics Heterogeneous Quantization