An official website of the United States government
Communication of model updates between client nodes and the central aggregating server is a major bottleneck in federated learning, especially in bandwidth-limited settings and high-dimensional models. Gradient quantization is an effective way of reducing the number of bits required to communicate each model update, albeit at the cost of having a higher error floor due to the higher variance of the stochastic gradients. In this work, we propose an adaptive quantization strategy called AdaQuantFL that aims to achieve communication efficiency as well as a low error floor by changing the number of quantization levels during the course of training. Experiments on training deep neural networks show that our method can converge in much fewer communicated bits as compared to fixed quantization level setups, with little or no impact on training and test accuracy.
more »
« less
This content will become publicly available on May 19, 2026
Does Quantization Improve Inference Speed? It Depends
Quantization is often cited as a technique for reducing model size and accelerating deep learning. However, past literature suggests that the effect of quantization on latency varies significantly across different settings, in some cases even increasing inference time rather than reducing it. To address this discrepancy, we conduct a series of systematic experiments on the Chameleon testbed to investigate the impact of three key variables on the effect of post-training quantization: the machine learning framework, the compute hardware, and the model itself. Our experiments demonstrate that each of these has a substantial impact on the overall inference time of a quantized model. Furthermore, we make experiment materials and artifacts publicly available so that others can validate our findings on the same hardware using Chameleon, and we share open educational resources on this topic that may be adopted in formal and informal education settings.
more »
« less
- Award ID(s):
- 2230079
- PAR ID:
- 10636811
- Publisher / Repository:
- IEEE
- Date Published:
- ISBN:
- 979-8-3315-0938-5
- Page Range / eLocation ID:
- 187-193
- Format(s):
- Medium: X
- Location:
- Tromsø, Norway
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
null (Ed.)Efficient machine learning implementations optimized for inference in hardware have wide-ranging benefits, depending on the application, from lower inference latency to higher data throughput and reduced energy consumption. Two popular techniques for reducing computation in neural networks are pruning, removing insignificant synapses, and quantization, reducing the precision of the calculations. In this work, we explore the interplay between pruning and quantization during the training of neural networks for ultra low latency applications targeting high energy physics use cases. Techniques developed for this study have potential applications across many other domains. We study various configurations of pruning during quantization-aware training, which we term quantization-aware pruning , and the effect of techniques like regularization, batch normalization, and different pruning schemes on performance, computational complexity, and information content metrics. We find that quantization-aware pruning yields more computationally efficient models than either pruning or quantization alone for our task. Further, quantization-aware pruning typically performs similar to or better in terms of computational efficiency compared to other neural architecture search techniques like Bayesian optimization. Surprisingly, while networks with different training configurations can have similar performance for the benchmark application, the information content in the network can vary significantly, affecting its generalizability.more » « less
-
Neural networks (NNs) have been extremely successful across many tasks in machine learning. Quantization of NN weights has become an important topic due to its impact on their energy efficiency, inference time and deployment on hardware. Although post-training quantization is well-studied, training optimal quantized NNs involves combinatorial non-convex optimization problems which appear intractable. In this work, we introduce a convex optimization strategy to train quantized NNs with polynomial activations. Our method leverages hidden convexity in twolayer neural networks from the recent literature, semidefinite lifting, and Grothendieck’s identity. Surprisingly, we show that certain quantized NN problems can be solved to global optimality provably in polynomial time in all relevant parameters via tight semidefinite relaxations. We present numerical examples to illustrate the effectiveness of our method.more » « less
-
Neural networks (NNs) have been extremely successful across many tasks in machine learning. Quantization of NN weights has become an important topic due to its impact on their energy efficiency, inference time and deployment on hardware. Although post-training quantization is well-studied, training optimal quantized NNs involves combinatorial non-convex optimization problems which appear intractable. In this work, we introduce a convex optimization strategy to train quantized NNs with polynomial activations. Our method leverages hidden convexity in two-layer neural networks from the recent literature, semidefinite lifting, and Grothendieck’s identity. Surprisingly, we show that certain quantized NN problems can be solved to global optimality provably in polynomial time in all relevant parameters via tight semidefinite relaxations. We present numerical examples to illustrate the effectiveness of our method.more » « less
-
This work targets the commonly used FPGA (field-programmable gate array) devices as the hardware platform for DNN edge computing. We focus on DNN quantization as the main model compression technique. The novelty of this work is: We use a quantization method that supports multiple precisions along the intra-layer dimension, while the existing quantization methods apply multi-precision quantization along the inter-layer dimension. The intra-layer multi-precision method can uniform the hardware configurations for different layers to reduce computation overhead and at the same time preserve the model accuracy as the inter-layer approach. Our proposed ILMPQ DNN quantization framework achieves 70.73% Top1 accuracy in ResNet-18 on the ImageNet dataset. We also validate the proposed MSP framework on two FPGA devices i.e., Xilinx XC7Z020 and XC7Z045. We achieve 3.65× speedup in end-to-end inference time on the ImageNet, comparing with the fixed-point quantization method.more » « less
