arXiv subjects

Yogish Sabharwal

Publications and source records attributed to Yogish Sabharwal.

At least 19 recordsLinked to original sources

arXiv2023

GREED: A Neural Framework for Learning Graph Distance Functions

Among various distance functions for graphs, graph and subgraph edit distances (GED and SED respectively) are two of the most popular and expressive measures. Unfortunately, exact computations for both are NP-hard. To overcome this computational bottleneck, neural approaches to learn and predict edit distance in polynomial time have received much interest. While considerable progress has been made, there exist limitations that need to be addressed. First, the efficacy of an approximate distance function lies not only in its approximation accuracy, but also in the preservation of its properties. To elaborate, although GED is a metric, its neural approximations do not provide such a guarantee. This prohibits their usage in higher order tasks that rely on metric distance functions, such as clustering or indexing. Second, several existing frameworks for GED do not extend to SED due to SED being asymmetric. In this work, we design a novel siamese graph neural network called GREED, which through a carefully crafted inductive bias, learns GED and SED in a property-preserving manner. Through extensive experiments across 10 real graph datasets containing up to 7 million edges, we establish that GREED is not only more accurate than the state of the art, but also up to 3 orders of magnitude faster. Even more significantly, due to preserving the triangle inequality, the generated embeddings are indexable and consequently, even in a CPU-only environment, GREED is up to 50 times faster than GPU-powered baselines for graph / subgraph retrieval.

cs.LG↗

arXiv2021

Rightsizing Clusters for Time-Limited Tasks

In conventional public clouds, designing a suitable initial cluster for a given application workload is important in reducing the computational foot-print during run-time. In edge or on-premise clouds, cold-start rightsizing the cluster at the time of installation is crucial in avoiding the recurrent capital expenditure. In both these cases, rightsizing has to balance cost-performance trade-off for a given application with multiple tasks, where each task can demand multiple resources, and the cloud offers nodes with different capacity and cost. Multidimensional bin-packing can address this cold-start rightsizing problem, but assumes that every task is always active. In contrast, real-world tasks (e.g. load bursts, batch and dead-lined tasks with time-limits) may be active only during specific time-periods or may have dynamic load profiles. The cluster cost can be reduced by reusing resources via time sharing and optimal packing. This motivates our generalized problem of cold-start rightsizing for time-limited tasks: given a timeline, time-periods and resource demands for tasks, the objective is to place the tasks on a minimum cost cluster of nodes without violating node capacities at any time instance. We design a baseline two-phase algorithm that performs penalty-based mapping of task to node-type and then, solves each node-type independently. We prove that the algorithm has an approximation ratio of O(D min(m, T)), where D, m and T are the number of resources, node-types and timeslots, respectively. We then present an improved linear programming based mapping strategy, enhanced further with a cross-node-type filling mechanism. Our experiments on synthetic and real-world cluster traces show significant cost reduction by LP-based mapping compared to the baseline, and the filling mechanism improves further to produce solutions within 20% of (a lower-bound to) the optimal solution.

cs.DC↗

arXiv2021

Generalizations of Length Limited Huffman Coding for Hierarchical Memory Settings

In this paper, we study the problem of designing prefix-free encoding schemes having minimum average code length that can be decoded efficiently under a decode cost model that captures memory hierarchy induced cost functions. We also study a special case of this problem that is closely related to the length limited Huffman coding (LLHC) problem; we call this the {\em soft-length limited Huffman coding} problem. In this version, there is a penalty associated with each of the $n$ characters of the alphabet whose encodings exceed a specified bound $D$($\leq n$), where the penalty increases linearly with the length of the encoding beyond $D$. The goal of the problem is to find a prefix-free encoding having minimum average code length and total penalty within a pre-specified bound ${\cal P}$. This generalizes the LLHC problem. We present an algorithm to solve this problem that runs in time $O( nD )$. We study a further generalization in which the penalty function and the objective function can both be arbitrary monotonically non-decreasing functions of the codeword length. We provide dynamic programming based exact and PTAS algorithms for this setting.

cs.DS↗

arXiv2021

Efficient Scaling of Dynamic Graph Neural Networks

We present distributed algorithms for training dynamic Graph Neural Networks (GNN) on large scale graphs spanning multi-node, multi-GPU systems. To the best of our knowledge, this is the first scaling study on dynamic GNN. We devise mechanisms for reducing the GPU memory usage and identify two execution time bottlenecks: CPU-GPU data transfer; and communication volume. Exploiting properties of dynamic graphs, we design a graph difference-based strategy to significantly reduce the transfer time. We develop a simple, but effective data distribution technique under which the communication volume remains fixed and linear in the input size, for any number of GPUs. Our experiments using billion-size graphs on a system of 128 GPUs shows that: (i) the distribution scheme achieves up to 30x speedup on 128 GPUs; (ii) the graph-difference technique reduces the transfer time by a factor of up to 4.1x and the overall execution time by up to 40%

cs.DC↗

arXiv2020

PoWER-BERT: Accelerating BERT Inference via Progressive Word-vector Elimination

We develop a novel method, called PoWER-BERT, for improving the inference time of the popular BERT model, while maintaining the accuracy. It works by: a) exploiting redundancy pertaining to word-vectors (intermediate encoder outputs) and eliminating the redundant vectors. b) determining which word-vectors to eliminate by developing a strategy for measuring their significance, based on the self-attention mechanism. c) learning how many word-vectors to eliminate by augmenting the BERT model and the loss function. Experiments on the standard GLUE benchmark shows that PoWER-BERT achieves up to 4.5x reduction in inference time over BERT with <1% loss in accuracy. We show that PoWER-BERT offers significantly better trade-off between accuracy and inference time compared to prior methods. We demonstrate that our method attains up to 6.8x reduction in inference time with <1% loss in accuracy when applied over ALBERT, a highly compressed version of BERT. The code for PoWER-BERT is publicly available at https://github.com/IBM/PoWER-BERT.

cs.LG↗

arXiv2020

Effective Elastic Scaling of Deep Learning Workloads

The increased use of deep learning (DL) in academia, government and industry has, in turn, led to the popularity of on-premise and cloud-hosted deep learning platforms, whose goals are to enable organizations utilize expensive resources effectively, and to share said resources among multiple teams in a fair and effective manner. In this paper, we examine the elastic scaling of Deep Learning (DL) jobs over large-scale training platforms and propose a novel resource allocation strategy for DL training jobs, resulting in improved job run time performance as well as increased cluster utilization. We begin by analyzing DL workloads and exploit the fact that DL jobs can be run with a range of batch sizes without affecting their final accuracy. We formulate an optimization problem that explores a dynamic batch size allocation to individual DL jobs based on their scaling efficiency, when running on multiple nodes. We design a fast dynamic programming based optimizer to solve this problem in real-time to determine jobs that can be scaled up/down, and use this optimizer in an autoscaler to dynamically change the allocated resources and batch sizes of individual DL jobs. We demonstrate empirically that our elastic scaling algorithm can complete up to $\approx 2 \times$ as many jobs as compared to a strong baseline algorithm that also scales the number of GPUs but does not change the batch size. We also demonstrate that the average completion time with our algorithm is up to $\approx 10 \times$ faster than that of the baseline.

cs.DC↗

arXiv2020

On Optimizing Distributed Tucker Decomposition for Sparse Tensors

The Tucker decomposition generalizes the notion of Singular Value Decomposition (SVD) to tensors, the higher dimensional analogues of matrices. We study the problem of constructing the Tucker decomposition of sparse tensors on distributed memory systems via the HOOI procedure, a popular iterative method. The scheme used for distributing the input tensor among the processors (MPI ranks) critically influences the HOOI execution time. Prior work has proposed different distribution schemes: an offline scheme based on sophisticated hypergraph partitioning method and simple, lightweight alternatives that can be used real-time. While the hypergraph based scheme typically results in faster HOOI execution time, being complex, the time taken for determining the distribution is an order of magnitude higher than the execution time of a single HOOI iteration. Our main contribution is a lightweight distribution scheme, which achieves the best of both worlds. We show that the scheme is near-optimal on certain fundamental metrics associated with the HOOI procedure and as a result, near-optimal on the computational load (FLOPs). Though the scheme may incur higher communication volume, the computation time is the dominant factor and as the result, the scheme achieves better performance on the overall HOOI execution time. Our experimental evaluation on large real-life tensors (having up to 4 billion elements) shows that the scheme outperforms the prior schemes on the HOOI execution time by a factor of up to 3x. On the other hand, its distribution time is comparable to the prior lightweight schemes and is typically lesser than the execution time of a single HOOI iteration.

cs.DC↗

arXiv2017

Efficient Training of Convolutional Neural Nets on Large Distributed Systems

Deep Neural Networks (DNNs) have achieved im- pressive accuracy in many application domains including im- age classification. Training of DNNs is an extremely compute- intensive process and is solved using variants of the stochastic gradient descent (SGD) algorithm. A lot of recent research has focussed on improving the performance of DNN training. In this paper, we present optimization techniques to improve the performance of the data parallel synchronous SGD algorithm using the Torch framework: (i) we maintain data in-memory to avoid file I/O overheads, (ii) we present a multi-color based MPI Allreduce algorithm to minimize communication overheads, and (iii) we propose optimizations to the Torch data parallel table framework that handles multi-threading. We evaluate the performance of our optimizations on a Power 8 Minsky cluster with 32 nodes and 128 NVidia Pascal P100 GPUs. With our optimizations, we are able to train 90 epochs of the ResNet-50 model on the Imagenet-1k dataset using 256 GPUs in just 48 minutes. This significantly improves on the previously best known performance of training 90 epochs of the ResNet-50 model on the same dataset using 256 GPUs in 65 minutes. To the best of our knowledge, this is the best known training performance demonstrated for the Imagenet- 1k dataset.

cs.DC↗

arXiv2017

Efficient Inferencing of Compressed Deep Neural Networks

Large number of weights in deep neural networks makes the models difficult to be deployed in low memory environments such as, mobile phones, IOT edge devices as well as "inferencing as a service" environments on cloud. Prior work has considered reduction in the size of the models, through compression techniques like pruning, quantization, Huffman encoding etc. However, efficient inferencing using the compressed models has received little attention, specially with the Huffman encoding in place. In this paper, we propose efficient parallel algorithms for inferencing of single image and batches, under various memory constraints. Our experimental results show that our approach of using variable batch size for inferencing achieves 15-25\% performance improvement in the inference throughput for AlexNet, while maintaining memory and latency constraints.

cs.DC↗

arXiv2017

Replica Placement on Bounded Treewidth Graphs

We consider the replica placement problem: given a graph with clients and nodes, place replicas on a minimum set of nodes to serve all the clients; each client is associated with a request and maximum distance that it can travel to get served and there is a maximum limit (capacity) on the amount of request a replica can serve. The problem falls under the general framework of capacitated set covering. It admits an O(\log n)-approximation and it is NP-hard to approximate within a factor of $o(\log n)$. We study the problem in terms of the treewidth $t$ of the graph and present an O(t)-approximation algorithm.

cs.DS↗

arXiv2017

On Optimizing Distributed Tucker Decomposition for Dense Tensors

The Tucker decomposition expresses a given tensor as the product of a small core tensor and a set of factor matrices. Apart from providing data compression, the construction is useful in performing analysis such as principal component analysis (PCA)and finds applications in diverse domains such as signal processing, computer vision and text analytics. Our objective is to develop an efficient distributed implementation for the case of dense tensors. The implementation is based on the HOOI (Higher Order Orthogonal Iterator) procedure, wherein the tensor-times-matrix product forms the core routine. Prior work have proposed heuristics for reducing the computational load and communication volume incurred by the routine. We study the two metrics in a formal and systematic manner, and design strategies that are optimal under the two fundamental metrics. Our experimental evaluation on a large benchmark of tensors shows that the optimal strategies provide significant reduction in load and volume compared to prior heuristics, and provide up to 7x speed-up in the overall running time.

cs.DC↗

arXiv2016

Improvable Knapsack Problems

We consider a variant of the knapsack problem, where items are available with different possible weights. Using a separate budget for these item improvements, the question is: Which items should be improved to which degree such that the resulting classic knapsack problem yields maximum profit? We present a detailed analysis for several cases of improvable knapsack problems, presenting constant factor approximation algorithms and two PTAS.

math.OC↗

arXiv2016

Subgraph Counting: Color Coding Beyond Trees

The problem of counting occurrences of query graphs in a large data graph, known as subgraph counting, is fundamental to several domains such as genomics and social network analysis. Many important special cases (e.g. triangle counting) have received significant attention. Color coding is a very general and powerful algorithmic technique for subgraph counting. Color coding has been shown to be effective in several applications, but scalable implementations are only known for the special case of {\em tree queries} (i.e. queries of treewidth one). In this paper we present the first efficient distributed implementation for color coding that goes beyond tree queries: our algorithm applies to any query graph of treewidth $2$. Since tree queries can be solved in time linear in the size of the data graph, our contribution is the first step into the realm of colour coding for queries that require superlinear running time in the worst case. This superlinear complexity leads to significant load balancing problems on graphs with heavy tailed degree distributions. Our algorithm structures the computation to work around high degree nodes in the data graph, and achieves very good runtime and scalability on a diverse collection of data and query graph pairs as a result. We also provide theoretical analysis of our algorithmic techniques, showing asymptotic improvements in runtime on random graphs with power law degree distributions, a popular model for real world graphs.

cs.DC↗

arXiv2013

Distributed and Parallel Algorithms for Set Cover Problems with Small Neighborhood Covers

In this paper, we study a class of set cover problems that satisfy a special property which we call the {\em small neighborhood cover} property. This class encompasses several well-studied problems including vertex cover, interval cover, bag interval cover and tree cover. We design unified distributed and parallel algorithms that can handle any set cover problem falling under the above framework and yield constant factor approximations. These algorithms run in polylogarithmic communication rounds in the distributed setting and are in NC, in the parallel setting.

cs.DS↗

arXiv2012

Scheduling Resources for Executing a Partial Set of Jobs

In this paper, we consider the problem of choosing a minimum cost set of resources for executing a specified set of jobs. Each input job is an interval, determined by its start-time and end-time. Each resource is also an interval determined by its start-time and end-time; moreover, every resource has a capacity and a cost associated with it. We consider two versions of this problem. In the partial covering version, we are also given as input a number k, specifying the number of jobs that must be performed. The goal is to choose k jobs and find a minimum cost set of resources to perform the chosen k jobs (at any point of time the capacity of the chosen set of resources should be sufficient to execute the jobs active at that time). We present an O(log n)-factor approximation algorithm for this problem. We also consider the prize collecting version, wherein every job also has a penalty associated with it. The feasible solution consists of a subset of the jobs, and a set of resources, to perform the chosen subset of jobs. The goal is to find a feasible solution that minimizes the sum of the costs of the selected resources and the penalties of the jobs that are not selected. We present a constant factor approximation algorithm for this problem

cs.DS↗

arXiv2012

Distributed Algorithms for Scheduling on Line and Tree Networks

We have a set of processors (or agents) and a set of graph networks defined over some vertex set. Each processor can access a subset of the graph networks. Each processor has a demand specified as a pair of vertices $ $, along with a profit; the processor wishes to send data between $u$ and $v$. Towards that goal, the processor needs to select a graph network accessible to it and a path connecting $u$ and $v$ within the selected network. The processor requires exclusive access to the chosen path, in order to route the data. Thus, the processors are competing for routes/channels. A feasible solution selects a subset of demands and schedules each selected demand on a graph network accessible to the processor owning the demand; the solution also specifies the paths to use for this purpose. The requirement is that for any two demands scheduled on the same graph network, their chosen paths must be edge disjoint. The goal is to output a solution having the maximum aggregate profit. Prior work has addressed the above problem in a distibuted setting for the special case where all the graph networks are simply paths (i.e, line-networks). Distributed constant factor approximation algorithms are known for this case. The main contributions of this paper are twofold. First we design a distributed constant factor approximation algorithm for the more general case of tree-networks. The core component of our algorithm is a tree-decomposition technique, which may be of independent interest. Secondly, for the case of line-networks, we improve the known approximation guarantees by a factor of 5. Our algorithms can also handle the capacitated scenario, wherein the demands and edges have bandwidth requirements and capacities, respectively.

cs.DS↗

arXiv2012

Mapping Strategies for the PERCS Architecture

The PERCS system was designed by IBM in response to a DARPA challenge that called for a high-productivity high-performance computing system. The IBM PERCS architecture is a two level direct network having low diameter and high bisection bandwidth. Mapping and routing strategies play an important role in the performance of applications on such a topology. In this paper, we study mapping strategies for PERCS architecture, that examine how to map tasks of a given job on to the physical processing nodes. We develop and present fundamental principles for designing good mapping strategies that minimize congestion. This is achieved via a theoretical study of some common communication patterns under both direct and indirect routing mechanisms supported by the architecture.

cs.DC↗

arXiv2012

Density Functions subject to a Co-Matroid Constraint

In this paper we consider the problem of finding the {\em densest} subset subject to {\em co-matroid constraints}. We are given a {\em monotone supermodular} set function $f$ defined over a universe $U$, and the density of a subset $S$ is defined to be $f(S)/\crd{S}$. This generalizes the concept of graph density. Co-matroid constraints are the following: given matroid $\calM$ a set $S$ is feasible, iff the complement of $S$ is {\em independent} in the matroid. Under such constraints, the problem becomes $\np$-hard. The specific case of graph density has been considered in literature under specific co-matroid constraints, for example, the cardinality matroid and the partition matroid. We show a 2-approximation for finding the densest subset subject to co-matroid constraints. Thus, for instance, we improve the approximation guarantees for the result for partition matroids in the literature.

cs.DS↗