arXiv subjects

Zigeng Wang

Publications and source records attributed to Zigeng Wang.

7 recordsLinked to original sources

arXiv2024

Zero-Space Cost Fault Tolerance for Transformer-based Language Models on ReRAM

Resistive Random Access Memory (ReRAM) has emerged as a promising platform for deep neural networks (DNNs) due to its support for parallel in-situ matrix-vector multiplication. However, hardware failures, such as stuck-at-fault defects, can result in significant prediction errors during model inference. While additional crossbars can be used to address these failures, they come with storage overhead and are not efficient in terms of space, energy, and cost. In this paper, we propose a fault protection mechanism that incurs zero space cost. Our approach includes: 1) differentiable structure pruning of rows and columns to reduce model redundancy, 2) weight duplication and voting for robust output, and 3) embedding duplicated most significant bits (MSBs) into the model weight. We evaluate our method on nine tasks of the GLUE benchmark with the BERT model, and experimental results prove its effectiveness.

cs.LG↗

arXiv2023

AutoReP: Automatic ReLU Replacement for Fast Private Network Inference

The growth of the Machine-Learning-As-A-Service (MLaaS) market has highlighted clients' data privacy and security issues. Private inference (PI) techniques using cryptographic primitives offer a solution but often have high computation and communication costs, particularly with non-linear operators like ReLU. Many attempts to reduce ReLU operations exist, but they may need heuristic threshold selection or cause substantial accuracy loss. This work introduces AutoReP, a gradient-based approach to lessen non-linear operators and alleviate these issues. It automates the selection of ReLU and polynomial functions to speed up PI applications and introduces distribution-aware polynomial approximation (DaPa) to maintain model expressivity while accurately approximating ReLUs. Our experimental results demonstrate significant accuracy improvements of 6.12% (94.31%, 12.9K ReLU budget, CIFAR-10), 8.39% (74.92%, 12.9K ReLU budget, CIFAR-100), and 9.45% (63.69%, 55K ReLU budget, Tiny-ImageNet) over current state-of-the-art methods, e.g., SNL. Morever, AutoReP is applied to EfficientNet-B2 on ImageNet dataset, and achieved 75.55% accuracy with 176.1 times ReLU budget reduction.

cs.CR↗

arXiv2022

Predicting band gaps and band-edge positions of oxide perovskites using DFT and machine learning

Density functional theory within the local or semilocal density approximations (DFT-LDA/GGA) has become a workhorse in electronic structure theory of solids, being extremely fast and reliable for energetics and structural properties, yet remaining highly inaccurate for predicting band gaps of semiconductors and insulators. Accurate prediction of band gaps using firstprinciples methods is time consuming, requiring hybrid functionals, quasi-particle GW, or quantum Monte Carlo methods. Efficiently correcting DFT-LDA/GGA band gaps and unveiling the main chemical and structural factors involved in this correction is desirable for discovering novel materials in high-throughput calculations. In this direction, we use DFT and machine learning techniques to correct band gaps and band-edge positions of a representative subset of ABO3 perovskite oxides. Relying on results of HSE06 hybrid functional calculations as target values of band gaps, we find a systematic band gap correction of ~1.5 eV for this class of materials, where ~1 eV comes from downward shifting the valence band and ~0.5 eV from uplifting the conduction band. The main chemical and structural factors determining the band gap correction are determined through a feature selection procedure.

cond-mat.mtrl-sci↗

arXiv2022

Deep Learning Based Page Creation for Improving E-Commerce Organic Search Traffic

Organic search comprises a large portion of the total traffic for e-commerce companies. One approach to expand company's exposure on organic search channel lies on creating landing pages having broader coverage on customer intentions. In this paper, we present a transformer language model based organic channel page management system aiming at increasing prominence of the company's overall clicks on the channel. Our system successfully handles the creation and deployment process of millions of new landing pages. We show and discuss the real-world performances of state-of-the-art language representation learning method, and reveal how we find them as the production-optimal solutions.

cs.CL↗

arXiv2021

An Efficient Group-based Search Engine Marketing System for E-Commerce

With the increasing scale of search engine marketing, designing an efficient bidding system is becoming paramount for the success of e-commerce companies. The critical challenges faced by a modern industrial-level bidding system include: 1. the catalog is enormous, and the relevant bidding features are of high sparsity; 2. the large volume of bidding requests induces significant computation burden to both the offline and online serving. Leveraging extraneous user-item information proves essential to mitigate the sparsity issue, for which we exploit the natural language signals from the users' query and the contextual knowledge from the products. In particular, we extract the vector representations of ads via the Transformer model and leverage their geometric relation to building collaborative bidding predictions via clustering. The two-step procedure also significantly reduces the computation stress of bid evaluation and optimization. In this paper, we introduce the end-to-end structure of the bidding system for search engine marketing for Walmart e-commerce, which successfully handles tens of millions of bids each day. We analyze the online and offline performances of our approach and discuss how we find it as a production-efficient solution.

cs.CL↗

arXiv2021

Against Membership Inference Attack: Pruning is All You Need

The large model size, high computational operations, and vulnerability against membership inference attack (MIA) have impeded deep learning or deep neural networks (DNNs) popularity, especially on mobile devices. To address the challenge, we envision that the weight pruning technique will help DNNs against MIA while reducing model storage and computational operation. In this work, we propose a pruning algorithm, and we show that the proposed algorithm can find a subnetwork that can prevent privacy leakage from MIA and achieves competitive accuracy with the original DNNs. We also verify our theoretical insights with experiments. Our experimental results illustrate that the attack accuracy using model compression is up to 13.6% and 10% lower than that of the baseline and Min-Max game, accordingly.

cs.LG↗

arXiv2018

Efficient Sequential and Parallel Algorithms for Estimating Higher Order Spectra

Polyspectral estimation is a problem of great importance in the analysis of nonlinear time series that has applications in biomedical signal processing, communications, geophysics, image, radar, sonar and speech processing, etc. Higher order spectra (HOS) have been used in unsupervised and supervised clustering in big data scenarios, in testing for Gaussianity, to suppress Gaussian noise, to characterize nonlinearities in time series data, and so on . Any algorithm for computing the $k$th order spectra of a time series of length $n$ needs $Ω(n^{k-1})$ time since the output size will be $Ω(n^{k-1})$ as well. Given that we live in an era of big data, $n$ could be very large. In this case, sequential algorithms might take unacceptable amounts of time. Thus it is essential to develop parallel algorithms. There is also room for improving existing sequential algorithms. In addition, parallel algorithms in the literature are nongeneric. In this paper we offer generic sequential algorithms for computing higher order spectra that are asymptotically faster than any published algorithm for HOS. Further, we offer memory efficient algorithms. We also present optimal parallel implementations of these algorithms on parallel computing models such as the PRAM and the mesh. We provide experimental results on our sequential and parallel algorithms. Our parallel implementation achieves very good speedups.

cs.DC↗