اشترك بالحزمة الذهبية واحصل على وصول غير محدود شمرا أكاديميا
تسجيل مستخدم جديدOptimal Communication-Computation Trade-Off in Heterogeneous Gradient Coding
371
0
0.0
(
0
)
نشر من قبل
Tayyebeh Jahani-Nezhad
تاريخ النشر
2021
مجال البحث
الهندسة المعلوماتية
والبحث باللغة
English
اسأل ChatGPT حول البحث
ﻻ يوجد ملخص باللغة العربية
Gradient coding allows a master node to derive the aggregate of the partial gradients, calculated by some worker nodes over the local data sets, with minimum communication cost, and in the presence of stragglers. In this paper, for gradient coding with linear encoding, we characterize the optimum communication cost for heterogeneous distributed systems with emph{arbitrary} data placement, with $s in mathbb{N}$ stragglers and $a in mathbb{N}$ adversarial nodes. In particular, we show that the optimum communication cost, normalized by the size of the gradient vectors, is equal to $(r-s-2a)^{-1}$, where $r in mathbb{N}$ is the minimum number that a data partition is replicated. In other words, the communication cost is determined by the data partition with the minimum replication, irrespective of the structure of the placement. The proposed achievable scheme also allows us to target the computation of a polynomial function of the aggregated gradient matrix. It also allows us to borrow some ideas from approximation computing and propose an approximate gradient coding scheme for the cases when the repetition in data placement is smaller than what is needed to meet the restriction imposed on communication cost or when the number of stragglers appears to be more than the presumed value in the system design.
قيم البحث
اقرأ أيضاً
The recent breakthrough in artificial intelligence (AI), especially deep neural networks (DNNs), has affected every branch of science and technology. Particularly, edge AI has been envisioned as a major application scenario to provide DNN-based servi
ces at edge devices. This article presents effective methods for edge inference at resource-constrained devices. It focuses on device-edge co-inference, assisted by an edge computing server, and investigates a critical trade-off among the computation cost of the on-device model and the communication cost of forwarding the intermediate feature to the edge server. A three-step framework is proposed for the effective inference: (1) model split point selection to determine the on-device model, (2) communication-aware model compression to reduce the on-device computation and the resulting communication overhead simultaneously, and (3) task-oriented encoding of the intermediate feature to further reduce the communication overhead. Experiments demonstrate that our proposed framework achieves a better trade-off and significantly reduces the inference latency than baseline methods.
In distributed machine learning (DML), the training data is distributed across multiple worker nodes to perform the underlying training in parallel. One major problem affecting the performance of DML algorithms is presence of stragglers. These are no
des that are terribly slow in performing their task which results in under-utilization of the training data that is stored in them. Towards this, gradient coding mitigates the impact of stragglers by adding sufficient redundancy in the data. Gradient coding and other straggler mitigation schemes assume that the straggler behavior of the worker nodes is identical. Our experiments on the Amazon AWS cluster however suggest otherwise and we see that there is a correlation in the straggler behavior across iterations. To model this, we introduce a heterogeneous straggler model where nodes are categorized into two classes, slow and active. To better utilize training data stored with slow nodes, we modify the existing gradient coding schemes with shuffling of the training data among workers. Our results (both simulation and cloud experiments) suggest remarkable improvement with shuffling over existing schemes. We perform theoretical analysis for the proposed models justifying their utility.
A central issue of distributed computing systems is how to optimally allocate computing and storage resources and design data shuffling strategies such that the total execution time for computing and data shuffling is minimized. This is extremely cri
tical when the computation, storage and communication resources are limited. In this paper, we study the resource allocation and coding scheme for the MapReduce-type framework with limited resources. In particular, we focus on the coded distributed computing (CDC) approach proposed by Li et al.. We first extend the asymmetric CDC (ACDC) scheme proposed by Yu et al. to the cascade case where each output function is computed by multiple servers. Then we demonstrate that whether CDC or ACDC is better depends on system parameters (e.g., number of computing servers) and task parameters (e.g., number of input files), implying that neither CDC nor ACDC is optimal. By merging the ideas of CDC and ACDC, we propose a hybrid scheme and show that it can strictly outperform CDC and ACDC. Furthermore, we derive an information-theoretic converse showing that for the MapReduce task using a type of weakly symmetric Reduce assignment, which includes the Reduce assignments of CDC and ACDC as special cases, the hybrid scheme with a corresponding resource allocation strategy is optimal, i.e., achieves the minimum execution time, for an arbitrary amount of computing servers and storage memories.
Distributed implementations of gradient-based methods, wherein a server distributes gradient computations across worker machines, need to overcome two limitations: delays caused by slow running machines called stragglers, and communication overheads.
Recently, Ye and Abbe [ICML 2018] proposed a coding-theoretic paradigm to characterize a fundamental trade-off between computation load per worker, communication overhead per worker, and straggler tolerance. However, their proposed coding schemes suffer from heavy decoding complexity and poor numerical stability. In this paper, we develop a communication-efficient gradient coding framework to overcome these drawbacks. Our proposed framework enables using any linear code to design the encoding and decoding functions. When a particular code is used in this framework, its block-length determines the computation load, dimension determines the communication overhead, and minimum distance determines the straggler tolerance. The flexibility of choosing a code allows us to gracefully trade-off the straggler threshold and communication overhead for smaller decoding complexity and higher numerical stability. Further, we show that using a maximum distance separable (MDS) code generated by a random Gaussian matrix in our framework yields a gradient code that is optimal with respect to the trade-off and, in addition, satisfies stronger guarantees on numerical stability as compared to the previously proposed schemes. Finally, we evaluate our proposed framework on Amazon EC2 and demonstrate that it reduces the average iteration time by 16% as compared to prior gradient coding schemes.
A major hurdle in machine learning is scalability to massive datasets. One approach to overcoming this is to distribute the computational tasks among several workers. textit{Gradient coding} has been recently proposed in distributed optimization to c
ompute the gradient of an objective function using multiple, possibly unreliable, worker nodes. By designing distributed coded schemes, gradient coded computations can be made resilient to textit{stragglers}, nodes with longer response time comparing to other nodes in a distributed network. Most such schemes rely on operations over the real or complex numbers and are inherently numerically unstable. We present a binary scheme which avoids such operations, thereby enabling numerically stable distributed computation of the gradient. Also, some restricting assumptions in prior work are dropped, and a more efficient decoding is given.
الأسئلة المقترحة
سجل دخول لتتمكن من نشر تعليقات
التعليقات
جاري جلب التعليقات
سجل دخول لتتمكن من متابعة معايير البحث التي قمت باختيارها
