Author name cluster

Ankur Agrawal

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

4 papers

2 author rows

NeurIPS Conference 2020 Conference Paper

Ultra-Low Precision 4-bit Training of Deep Neural Networks

Xiao Sun
Naigang Wang
Chia-Yu Chen
Jiamin Ni
Ankur Agrawal
Xiaodong Cui
Swagath Venkataramani
Kaoutar El Maghraoui

In this paper, we propose a number of novel techniques and numerical representation formats that enable, for the very first time, the precision of training systems to be aggressively scaled from 8-bits to 4-bits. To enable this advance, we explore a novel adaptive Gradient Scaling technique (Gradscale) that addresses the challenges of insufficient range and resolution in quantized gradients as well as explores the impact of quantization errors observed during model training. We theoretically analyze the role of bias in gradient quantization and propose solutions that mitigate the impact of this bias on model convergence. Finally, we examine our techniques on a spectrum of deep learning models in computer vision, speech, and NLP. In combination with previously proposed solutions for 4-bit quantization of weight and activation tensors, 4-bit training shows a non-significant loss in accuracy across application domains while enabling significant hardware acceleration (> 7X over state-of-the-art FP16 systems).

PDF Details

AAAI Conference 2018 Conference Paper

AdaComp: Adaptive Residual Gradient Compression for Data-Parallel Distributed Training

Chia-Yu Chen
Jungwook Choi
Daniel Brand
Ankur Agrawal
Wei Zhang
Kailash Gopalakrishnan

Highly distributed training of Deep Neural Networks (DNNs) on future compute platforms (offering 100 of TeraOps/s of computational capacity) is expected to be severely communication constrained. To overcome this limitation, new gradient compression techniques are needed that are computationally friendly, applicable to a wide variety of layers seen in Deep Neural Networks and adaptable to variations in network architectures as well as their hyper-parameters. In this paper we introduce a novel technique - the Adaptive Residual Gradient Compression (AdaComp) scheme. AdaComp is based on localized selection of gradient residues and automatically tunes the compression rate depending on local activity. We show excellent results on a wide spectrum of state of the art Deep Learning models in multiple domains (vision, speech, language), datasets (MNIST, CIFAR10, ImageNet, BN50, Shakespeare), optimizers (SGD with momentum, Adam) and network parameters (number of learners, minibatch-size etc.). Exploiting both sparsity and quantization, we demonstrate end-to-end compression rates of ∼200× for fully-connected and recurrent layers, and ∼40× for convolutional layers, without any noticeable degradation in model accuracies.

PDF Details

ICML Conference 2015 Conference Paper

Deep Learning with Limited Numerical Precision

Suyog Gupta
Ankur Agrawal
Kailash Gopalakrishnan
Pritish Narayanan

Training of large-scale deep neural networks is often constrained by the available computational resources. We study the effect of limited precision data representation and computation on neural network training. Within the context of low-precision fixed-point computations, we observe the rounding scheme to play a crucial role in determining the network’s behavior during training. Our results show that deep networks can be trained using only 16-bit wide fixed-point number representation when using stochastic rounding, and incur little to no degradation in the classification accuracy. We also demonstrate an energy-efficient hardware accelerator that implements low-precision fixed-point arithmetic with stochastic rounding

Details

AIIM Journal 2013 Journal Article

The readiness of SNOMED problem list concepts for meaningful use of electronic health records

Ankur Agrawal
Zhe He
Yehoshua Perl
Duo Wei
Michael Halper
Gai Elhanan
Yan Chen

Objective By 2015, SNOMED CT (SCT) will become the USA's standard for encoding diagnoses and problem lists in electronic health records (EHRs). To facilitate this effort, the National Library of Medicine has published the “SCT Clinical Observations Recording and Encoding” and the “Veterans Health Administration and Kaiser Permanente” problem lists (collectively, the “PL”). The PL is studied in regard to its readiness to support meaningful use of EHRs. In particular, we wish to determine if inconsistencies appearing in SCT, in general, occur as frequently in the PL, and whether further quality-assurance (QA) efforts on the PL are required. Methods and materials A study is conducted where two random samples of SCT concepts are compared. The first consists of concepts strictly from the PL and the second contains general SCT concepts distributed proportionally to the PL's in terms of their hierarchies. Each sample is analyzed for its percentage of primitive concepts and for frequency of modeling errors of various severity levels as quality measures. A simple structural indicator, namely, the number of parents, is suggested to locate high likelihood inconsistencies in hierarchical relationships. The effectiveness of this indicator is evaluated. Results PL concepts are found to be slightly better than other concepts in the respective SCT hierarchies with regards to the quality measure of the percentage of primitive concepts and the frequency of modeling errors. There were 58% primitive concepts in the PL sample versus 62% in the control sample. The structural indicator of number of parents is shown to be statistically significant in its ability to identify concepts having a higher likelihood of inconsistencies in their hierarchical relationships. The absolute number of errors in the group of concepts having 1–3 parents was shown to be significantly lower than that for concepts with 4–6 parents and those with 7 or more parents based on Chi-squared analyses. Conclusion PL concepts suffer from the same issues as general SCT concepts, although to a slightly lesser extent, and do require further QA efforts to promote meaningful use of EHRs. To support such efforts, a structural indicator is shown to effectively ferret out potentially problematic concepts where those QA efforts should be focused.

Details DOI