arXiv subjects

Xi Xie

Publications and source records attributed to Xi Xie.

24 records · Page 2Linked to original sources

arXiv2023

Accel-GCN: High-Performance GPU Accelerator Design for Graph Convolution Networks

Graph Convolutional Networks (GCNs) are pivotal in extracting latent information from graph data across various domains, yet their acceleration on mainstream GPUs is challenged by workload imbalance and memory access irregularity. To address these challenges, we present Accel-GCN, a GPU accelerator architecture for GCNs. The design of Accel-GCN encompasses: (i) a lightweight degree sorting stage to group nodes with similar degree; (ii) a block-level partition strategy that dynamically adjusts warp workload sizes, enhancing shared memory locality and workload balance, and reducing metadata overhead compared to designs like GNNAdvisor; (iii) a combined warp strategy that improves memory coalescing and computational parallelism in the column dimension of dense matrices. Utilizing these principles, we formulated a kernel for sparse matrix multiplication (SpMM) in GCNs that employs block-level partitioning and combined warp strategy. This approach augments performance and multi-level memory efficiency and optimizes memory bandwidth by exploiting memory coalescing and alignment. Evaluation of Accel-GCN across 18 benchmark graphs reveals that it outperforms cuSPARSE, GNNAdvisor, and graph-BLAST by factors of 1.17 times, 1.86 times, and 2.94 times respectively. The results underscore Accel-GCN as an effective solution for enhancing GCN computational efficiency.

cs.AR↗

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↗

arXiv2023

RRNet: Towards ReLU-Reduced Neural Network for Two-party Computation Based Private Inference

The proliferation of deep learning (DL) has led to the emergence of privacy and security concerns. To address these issues, secure Two-party computation (2PC) has been proposed as a means of enabling privacy-preserving DL computation. However, in practice, 2PC methods often incur high computation and communication overhead, which can impede their use in large-scale systems. To address this challenge, we introduce RRNet, a systematic framework that aims to jointly reduce the overhead of MPC comparison protocols and accelerate computation through hardware acceleration. Our approach integrates the hardware latency of cryptographic building blocks into the DNN loss function, resulting in improved energy efficiency, accuracy, and security guarantees. Furthermore, we propose a cryptographic hardware scheduler and corresponding performance model for Field Programmable Gate Arrays (FPGAs) to further enhance the efficiency of our framework. Experiments show RRNet achieved a much higher ReLU reduction performance than all SOTA works on CIFAR-10 dataset.

cs.CR↗

arXiv2022

On the Niho type locally-APN power functions and their boomerang spectrum

In this article, we focus on the concept of locally-APN-ness (``APN" is the abbreviation of the well-known notion of Almost Perfect Nonlinear) introduced by Blondeau, Canteaut, and Charpin, which makes the corpus of S-boxes somehow larger regarding their differential uniformity and, therefore, possibly, more suitable candidates against the differential attack (or their variants). Specifically, given two coprime positive integers $m$ and $k$ such that $\gcd(2^m+1,2^k+1)=1$, we investigate the locally-APN-ness property of an infinite family of Niho type power functions in the form $F(x)=x^{s(2^m-1)+1}$ over the finite field ${\mathbb F}_{2^{2m}}$ for $s=(2^k+1)^{-1}$, where $(2^k+1)^{-1}$ denotes the multiplicative inverse modulo $2^m+1$. By employing finer studies of the number of solutions of certain equations over finite fields (with even characteristic) as well as some subtle manipulations of solving some equations, we prove that $F(x)$ is locally APN and determine its differential spectrum. It is worth noting that computer experiments show that this class of locally-APN power functions covers all Niho type locally-APN power functions for $2\leq m\leq10$. In addition, we also determine the boomerang spectrum of $F(x)$ by using its differential spectrum, which particularly generalizes a recent result by Yan, Zhang, and Li.

cs.IT↗

arXiv2021

Graphene Based Opt-Thermoelectric Tweezers

Since the discovery of graphene, its excellent physical properties has greatly improved the performance of many optoelectronic devices and brought important technological revolution to optical research and application. Here, we introduce graphene into the field of optical tweezers technology and demonstrate a new thermoelectric optical tweezers technology based on graphene. This technology can not only reduce the incident light energy by 2 orders of magnitude (compared with traditional optical tweezers), but also bring new advantages such as much broader working bandwidth and larger working area than the thermoelectric optical tweezers based on gold film widely studied before. Compared with gold film, graphene has more novel characteristics like high thermal conductivity, high uniformity and easy process. Thus, we found even monolayer graphene can achieve stable trapping for particles in a broad band, and the performance is enhanced with more graphene layers. Furthermore, structured graphene patterns can be easily generated to holographically trap multiple particles as desired shapes. This work verifies the great application potential of two-dimensional materials in op-tical tweezers technology, and it will promote more promising applications in cell trapping, tapping or concentration of biomolecules, microfluidics and biosensors.

physics.optics↗

arXiv2021

Several classes of bent functions over finite fields

Let $\mathbb{F}_{p^{n}}$ be the finite field with $p^n$ elements and $\operatorname{Tr}(\cdot)$ be the trace function from $\mathbb{F}_{p^{n}}$ to $\mathbb{F}_{p}$, where $p$ is a prime and $n$ is an integer. Inspired by the works of Mesnager (IEEE Trans. Inf. Theory 60(7): 4397-4407, 2014) and Tang et al. (IEEE Trans. Inf. Theory 63(10): 6149-6157, 2017), we study a class of bent functions of the form $f(x)=g(x)+F(\operatorname{Tr}(u_1x),\operatorname{Tr}(u_2x),\cdots,\operatorname{Tr}(u_τx))$, where $g(x)$ is a function from $\mathbb{F}_{p^{n}}$ to $\mathbb{F}_{p}$, $τ\geq2$ is an integer, $F(x_1,\cdots,x_n)$ is a reduced polynomial in $\mathbb{F}_{p}[x_1,\cdots,x_n]$ and $u_i\in \mathbb{F}^{*}_{p^n}$ for $1\leq i \leq τ$. As a consequence, we obtain a generic result on the Walsh transform of $f(x)$ and characterize the bentness of $f(x)$ when $g(x)$ is bent for $p=2$ and $p>2$ respectively. Our results generalize some earlier works. In addition, we study the construction of bent functions $f(x)$ when $g(x)$ is not bent for the first time and present a class of bent functions from non-bent Gold functions.

cs.IT↗