arXiv subjects

Sian Jin

Publications and source records attributed to Sian Jin.

41 records · Page 3Linked to original sources

arXiv2019

Optimization-based Decentralized Coded Caching for Files and Caches with Arbitrary Size

Existing decentralized coded caching solutions cannot guarantee small loads in the general scenario with arbitrary file sizes and cache sizes. In this paper, we propose an optimization framework for decentralized coded caching in the general scenario to minimize the worst-case load and average load (under an arbitrary file popularity), respectively. Specifically, we first propose a class of decentralized coded caching schemes for the general scenario, which are specified by a general caching parameter and include several known schemes as special cases. Then, we optimize the caching parameter to minimize the worst-case load and average load, respectively. Each of the two optimization problems is a challenging nonconvex problem with a nondifferentiable objective function. For each optimization problem, we develop an iterative algorithm to obtain a stationary point using techniques for solving Complementary Geometric Programming (GP). We also obtain a low-complexity approximate solution by solving an approximate problem with a differentiable objective function which is an upper bound on the original nondifferentiable one, and characterize the performance loss caused by the approximation. Finally, we present two information-theoretic converse bounds on the worst-case load and average load (under an arbitrary file popularity) in the general scenario, respectively. To the best of our knowledge, this is the first work that provides optimization-based decentralized coded caching schemes and information-theoretic converse bounds for the general scenario.

cs.IT↗

arXiv2019

DeepSZ: A Novel Framework to Compress Deep Neural Networks by Using Error-Bounded Lossy Compression

DNNs have been quickly and broadly exploited to improve the data analysis quality in many complex science and engineering applications. Today's DNNs are becoming deeper and wider because of increasing demand on the analysis quality and more and more complex applications to resolve. The wide and deep DNNs, however, require large amounts of resources, significantly restricting their utilization on resource-constrained systems. Although some network simplification methods have been proposed to address this issue, they suffer from either low compression ratios or high compression errors, which may introduce a costly retraining process for the target accuracy. In this paper, we propose DeepSZ: an accuracy-loss bounded neural network compression framework, which involves four key steps: network pruning, error bound assessment, optimization for error bound configuration, and compressed model generation, featuring a high compression ratio and low encoding time. The contribution is three-fold. (1) We develop an adaptive approach to select the feasible error bounds for each layer. (2) We build a model to estimate the overall loss of accuracy based on the accuracy degradation caused by individual decompressed layers. (3) We develop an efficient optimization algorithm to determine the best-fit configuration of error bounds in order to maximize the compression ratio under the user-set accuracy constraint. Experiments show that DeepSZ can compress AlexNet and VGG-16 on the ImageNet by a compression ratio of 46X and 116X, respectively, and compress LeNet-300-100 and LeNet-5 on the MNIST by a compression ratio of 57X and 56X, respectively, with only up to 0.3% loss of accuracy. Compared with other state-of-the-art methods, DeepSZ can improve the compression ratio by up to 1.43X, the DNN encoding performance by up to 4.0X (with four Nvidia Tesla V100 GPUs), and the decoding performance by up to 6.2X.

cs.CV↗

arXiv2018

Uncoded Placement Optimization for Coded Delivery

We consider the classical coded caching problem as defined by Maddah-Ali and Niesen, where a server with a library of $N$ files of equal size is connected to $K$ users via a shared error-free link. Each user is equipped with a cache with capacity of $M$ files. The goal is to design a static content placement and delivery scheme such that the average load over the shared link is minimized. We first present a class of centralized coded caching schemes consisting of a general content placement strategy specified by a file partition parameter, enabling efficient and flexible content placement, and a specific content delivery strategy, enabling load reduction by exploiting common requests of different users. For the proposed class of schemes, we consider two cases for the optimization of the file partition parameter, depending on whether a large subpacketization level is allowed or not. In the case of an unrestricted subpacketization level, we formulate the coded caching optimization in order to minimize the average load under an arbitrary file popularity. A direct formulation of the problem involves $N2^K$ variables. By imposing some additional conditions, the problem is reduced to a linear program with $N(K+1)$ variables under an arbitrary file popularity and with $K+1$ variables under the uniform file popularity. We can recover Yu {\em et al.}'s optimal scheme for the uniform file popularity as an optimal solution of our problem. When a low subpacketization level is desired, we introduce a subpacketization level constraint involving the $\ell_0$ norm for each file. Again, by imposing the same additional conditions, we can simplify the problem to a difference of two convex functions (DC) problem with $N(K+1)$ variables that can be efficiently solved.

cs.IT↗

arXiv2017

Structural Properties of Uncoded Placement Optimization for Coded Delivery

A centralized coded caching scheme has been proposed by Maddah-Ali and Niesen to reduce the worst-case load of a network consisting of a server with access to N files and connected through a shared link to K users, each equipped with a cache of size M. However, this centralized coded caching scheme is not able to take advantage of a non-uniform, possibly very skewed, file popularity distribution. In this work, we consider the same network setting but aim to reduce the average load under an arbitrary (known) file popularity distribution. First, we consider a class of centralized coded caching schemes utilizing general uncoded placement and a specific coded delivery strategy, which are specified by a general file partition parameter. Then, we formulate the coded caching design optimization problem over the considered class of schemes with 2^K2^N variables to minimize the average load by optimizing the file partition parameter under an arbitrary file popularity. Furthermore, we show that the optimization problem is convex, and the resulting optimal solution generally improves upon known schemes. Next, we analyze structural properties of the optimization problem to obtain design insights and reduce the complexity. Specifically, we obtain an equivalent linear optimization problem with (K+1)N variables under an arbitrary file popularity and an equivalent linear optimization problem with K+1 variables under the uniform file popularity. Under the uniform file popularity, we also obtain the closed form optimal solution, which corresponds to Maddah-Ali-Niesen's centralized coded caching scheme. Finally, we present an information-theoretic converse bound on the average load under an arbitrary file popularity.

cs.IT↗

arXiv2017

New Order-Optimal Decentralized Coded Caching Schemes with Good Performance in the Finite File Size Regime

A decentralized coded caching scheme based on independent random content placement has been proposed by Maddah-Ali and Niesen, and has been shown to achieve an order-optimal memory-load tradeoff when the file size goes to infinity. It was then successively shown by Shanmugam et al. that in the practical operating regime where the file size is limited such scheme yields much less attractive load gain. In this paper, we propose a decentralized random coded caching scheme and a partially decentralized sequential coded caching scheme with different coordination requirements in the content placement phase. The proposed content placement and delivery methods aim at ensuring abundant coded-multicasting opportunities in the content delivery phase when the file size is finite. We first analyze the loads of the two proposed schemes and show that the sequential scheme outperforms the random scheme in the finite file size regime. We also show that both our proposed schemes outperform Maddah-Ali--Niesen's and Shanmugam et al.'s decentralized schemes for finite file size, when the number of users is sufficiently large. Then, we show that our schemes achieve the same memory-load tradeoff as Maddah-Ali--Niesen's decentralized scheme when the file size goes to infinity, and hence are also order optimal. Finally, we analyze the load gains of the two proposed schemes and characterize the corresponding required file sizes.

cs.IT↗