arXiv subjects

James Demmel

Publications and source records attributed to James Demmel.

At least 37 records · Page 2Linked to original sources

arXiv2024

Distributed-Memory Randomized Algorithms for Sparse Tensor CP Decomposition

Candecomp / PARAFAC (CP) decomposition, a generalization of the matrix singular value decomposition to higher-dimensional tensors, is a popular tool for analyzing multidimensional sparse data. On tensors with billions of nonzero entries, computing a CP decomposition is a computationally intensive task. We propose the first distributed-memory implementations of two randomized CP decomposition algorithms, CP-ARLS-LEV and STS-CP, that offer nearly an order-of-magnitude speedup at high decomposition ranks over well-tuned non-randomized decomposition packages. Both algorithms rely on leverage score sampling and enjoy strong theoretical guarantees, each with varying time and accuracy tradeoffs. We tailor the communication schedule for our random sampling algorithms, eliminating expensive reduction collectives and forcing communication costs to scale with the random sample count. Finally, we optimize the local storage format for our methods, switching between analogues of compressed sparse column and compressed sparse row formats. Experiments show that our methods are fast and scalable, producing 11x speedup over SPLATT by decomposing the billion-scale Reddit tensor on 512 CPU cores in under two minutes.

math.NA↗

arXiv2024

Fast Exact Leverage Score Sampling from Khatri-Rao Products with Applications to Tensor Decomposition

We present a data structure to randomly sample rows from the Khatri-Rao product of several matrices according to the exact distribution of its leverage scores. Our proposed sampler draws each row in time logarithmic in the height of the Khatri-Rao product and quadratic in its column count, with persistent space overhead at most the size of the input matrices. As a result, it tractably draws samples even when the matrices forming the Khatri-Rao product have tens of millions of rows each. When used to sketch the linear least squares problems arising in CANDECOMP / PARAFAC tensor decomposition, our method achieves lower asymptotic complexity per solve than recent state-of-the-art methods. Experiments on billion-scale sparse tensors validate our claims, with our algorithm achieving higher accuracy than competing methods as the decomposition rank grows.

math.NA↗

arXiv2023

Computron: Serving Distributed Deep Learning Models with Model Parallel Swapping

Many of the most performant deep learning models today in fields like language and image understanding are fine-tuned models that contain billions of parameters. In anticipation of workloads that involve serving many of such large models to handle different tasks, we develop Computron, a system that uses memory swapping to serve multiple distributed models on a shared GPU cluster. Computron implements a model parallel swapping design that takes advantage of the aggregate CPU-GPU link bandwidth of a cluster to speed up model parameter transfers. This design makes swapping large models feasible and can improve resource utilization. We demonstrate that Computron successfully parallelizes model swapping on multiple GPUs, and we test it on randomized workloads to show how it can tolerate real world variability factors like burstiness and skewed request rates. Computron's source code is available at https://github.com/dlzou/computron.

cs.DC↗

arXiv2023

Randomized Numerical Linear Algebra : A Perspective on the Field With an Eye to Software

Randomized numerical linear algebra - RandNLA, for short - concerns the use of randomization as a resource to develop improved algorithms for large-scale linear algebra computations. The origins of contemporary RandNLA lay in theoretical computer science, where it blossomed from a simple idea: randomization provides an avenue for computing approximate solutions to linear algebra problems more efficiently than deterministic algorithms. This idea proved fruitful in the development of scalable algorithms for machine learning and statistical data analysis applications. However, RandNLA's true potential only came into focus upon integration with the fields of numerical analysis and "classical" numerical linear algebra. Through the efforts of many individuals, randomized algorithms have been developed that provide full control over the accuracy of their solutions and that can be every bit as reliable as algorithms that might be found in libraries such as LAPACK. Recent years have even seen the incorporation of certain RandNLA methods into MATLAB, the NAG Library, NVIDIA's cuSOLVER, and SciKit-Learn. For all its success, we believe that RandNLA has yet to realize its full potential. In particular, we believe the scientific community stands to benefit significantly from suitably defined "RandBLAS" and "RandLAPACK" libraries, to serve as standards conceptually analogous to BLAS and LAPACK. This 200-page monograph represents a step toward defining such standards. In it, we cover topics spanning basic sketching, least squares and optimization, low-rank approximation, full matrix decompositions, leverage score sampling, and sketching data with tensor product structures (among others). Much of the provided pseudo-code has been tested via publicly available MATLAB and Python implementations.

math.NA↗

arXiv2022

Proposed Consistent Exception Handling for the BLAS and LAPACK

Numerical exceptions, which may be caused by overflow, operations like division by 0 or sqrt(-1), or convergence failures, are unavoidable in many cases, in particular when software is used on unforeseen and difficult inputs. As more aspects of society become automated, e.g., self-driving cars, health monitors, and cyber-physical systems more generally, it is becoming increasingly important to design software that is resilient to exceptions, and that responds to them in a consistent way. Consistency is needed to allow users to build higher-level software that is also resilient and consistent (and so on recursively). In this paper we explore the design space of consistent exception handling for the widely used BLAS and LAPACK linear algebra libraries, pointing out a variety of instances of inconsistent exception handling in the current versions, and propose a new design that balances consistency, complexity, ease of use, and performance. Some compromises are needed, because there are preexisting inconsistencies that are outside our control, including in or between existing vendor BLAS implementations, different programming languages, and even compilers for the same programming language. And user requests from our surveys are quite diverse. We also propose our design as a possible model for other numerical software, and welcome comments on our design choices.

cs.MS↗

arXiv2022

Communication Bounds for Convolutional Neural Networks

Convolutional neural networks (CNNs) are important in a wide variety of machine learning tasks and applications, so optimizing their performance is essential. Moving words of data between levels of a memory hierarchy or between processors on a network is much more expensive than the cost of arithmetic, so minimizing communication is critical to optimizing performance. In this paper, we present new lower bounds on data movement for mixed precision convolutions in both single-processor and parallel distributed memory models, as well as algorithms that outperform current implementations such as Im2Col. We obtain performance figures using GEMMINI, a machine learning accelerator, where our tiling provides improvements between 13% and 150% over a vendor supplied algorithm.

cs.DC↗

arXiv2022

Distributed-Memory Sparse Kernels for Machine Learning

Sampled Dense Times Dense Matrix Multiplication (SDDMM) and Sparse Times Dense Matrix Multiplication (SpMM) appear in diverse settings, such as collaborative filtering, document clustering, and graph embedding. Frequently, the SDDMM output becomes the input sparse matrix for a subsequent SpMM operation. Existing work has focused on shared memory parallelization of these primitives. While there has been extensive analysis of communication-minimizing distributed 1.5D algorithms for SpMM, no such analysis exists for SDDMM or the back-to-back sequence of SDDMM and SpMM, termed FusedMM. We show that distributed memory 1.5D and 2.5D algorithms for SpMM can be converted to algorithms for SDDMM with identical communication costs and input / output data layouts. Further, we give two communication-eliding strategies to reduce costs further for FusedMM kernels: either reusing the replication of an input dense matrix for the SDDMM and SpMM in sequence, or fusing the local SDDMM and SpMM kernels. We benchmark FusedMM algorithms on Cori, a Cray XC40 at LBNL, using Erdos-Renyi random matrices and large real-world sparse matrices. On 256 nodes with 68 cores each, 1.5D FusedMM algorithms using either communication eliding approach can save at least 30% of time spent exclusively in communication compared to executing a distributed-memory SpMM and SDDMM kernel in sequence. On real-world matrices with hundreds of millions of edges, all of our algorithms exhibit at least a 10x speedup over the SpMM algorithm in PETSc. On these matrices, our communication-eliding techniques exhibit runtimes up to 1.6 times faster than an unoptimized sequence of SDDMM and SpMM. We embed and test the scaling of our algorithms in real-world applications, including collaborative filtering via alternating-least-squares and inference for attention-based graph neural networks.

cs.DC↗

arXiv2021

Communication Lower Bounds of Bilinear Algorithms for Symmetric Tensor Contractions

We introduce a new theoretical framework for deriving lower bounds on data movement in bilinear algorithms. Bilinear algorithms are a general representation of fast algorithms for bilinear functions, which include computation of matrix multiplication, convolution, and symmetric tensor contractions. A bilinear algorithm is described by three matrices. Our communication lower bounds are based on quantifying the minimal matrix ranks of matching subsets of columns of these matrices. This infrastructure yields new communication lower bounds for symmetric tensor contraction algorithms, which provide qualitative new insights. Tensor symmetry (invariance under permutation of modes) is common to many applications of tensor computations (e.g., tensor representation of hypergraphs, analysis of high order moments in data, as well as tensors modelling interactions of electrons in computational chemistry). Tensor symmetry enables reduction in representation size as well as arithmetic cost of contractions by factors that scale with the number of equivalent permutations. However, we derive lower bounds showing that these arithmetic cost and memory reductions can necessitate increases in data movement by factors that scale with the size of the tensors.

cs.DC↗

arXiv2021

Auto-Precision Scaling for Distributed Deep Learning

It has been reported that the communication cost for synchronizing gradients can be a bottleneck, which limits the scalability of distributed deep learning. Using low-precision gradients is a promising technique for reducing the bandwidth requirement. In this work, we propose Auto Precision Scaling (APS), an algorithm that can improve the accuracy when we communicate gradients by low-precision floating-point values. APS can improve the accuracy for all precisions with a trivial communication cost. Our experimental results show that for many applications, APS can train state-of-the-art models by 8-bit gradients with no or only a tiny accuracy loss (<0.05%). Furthermore, we can avoid any accuracy loss by designing a hybrid-precision technique. Finally, we propose a performance model to evaluate the proposed method. Our experimental results show that APS can get a significant speedup over state-of-the-art methods. To make it available to researchers and developers, we design and implement CPD (Customized-Precision Deep Learning) system, which can simulate the training process using an arbitrary low-precision customized floating-point format. We integrate CPD into PyTorch and make it open-source.

cs.DC↗

arXiv2021

CoSA: Scheduling by Constrained Optimization for Spatial Accelerators

Recent advances in Deep Neural Networks (DNNs) have led to active development of specialized DNN accelerators, many of which feature a large number of processing elements laid out spatially, together with a multi-level memory hierarchy and flexible interconnect. While DNN accelerators can take advantage of data reuse and achieve high peak throughput, they also expose a large number of runtime parameters to the programmers who need to explicitly manage how computation is scheduled both spatially and temporally. In fact, different scheduling choices can lead to wide variations in performance and efficiency, motivating the need for a fast and efficient search strategy to navigate the vast scheduling space. To address this challenge, we present CoSA, a constrained-optimization-based approach for scheduling DNN accelerators. As opposed to existing approaches that either rely on designers' heuristics or iterative methods to navigate the search space, CoSA expresses scheduling decisions as a constrained-optimization problem that can be deterministically solved using mathematical optimization techniques. Specifically, CoSA leverages the regularities in DNN operators and hardware to formulate the DNN scheduling space into a mixed-integer programming (MIP) problem with algorithmic and architectural constraints, which can be solved to automatically generate a highly efficient schedule in one shot. We demonstrate that CoSA-generated schedules significantly outperform state-of-the-art approaches by a geometric mean of up to 2.5x across a wide range of DNN networks while improving the time-to-solution by 90x.

cs.LG↗

arXiv2020

Avoiding Communication in Logistic Regression

Stochastic gradient descent (SGD) is one of the most widely used optimization methods for solving various machine learning problems. SGD solves an optimization problem by iteratively sampling a few data points from the input data, computing gradients for the selected data points, and updating the solution. However, in a parallel setting, SGD requires interprocess communication at every iteration. We introduce a new communication-avoiding technique for solving the logistic regression problem using SGD. This technique re-organizes the SGD computations into a form that communicates every $s$ iterations instead of every iteration, where $s$ is a tuning parameter. We prove theoretical flops, bandwidth, and latency upper bounds for SGD and its new communication-avoiding variant. Furthermore, we show experimental results that illustrate that the new Communication-Avoiding SGD (CA-SGD) method can achieve speedups of up to $4.97\times$ on a high-performance Infiniband cluster without altering the convergence behavior or accuracy.

cs.LG↗

arXiv2020

Training EfficientNets at Supercomputer Scale: 83% ImageNet Top-1 Accuracy in One Hour

EfficientNets are a family of state-of-the-art image classification models based on efficiently scaled convolutional neural networks. Currently, EfficientNets can take on the order of days to train; for example, training an EfficientNet-B0 model takes 23 hours on a Cloud TPU v2-8 node. In this paper, we explore techniques to scale up the training of EfficientNets on TPU-v3 Pods with 2048 cores, motivated by speedups that can be achieved when training at such scales. We discuss optimizations required to scale training to a batch size of 65536 on 1024 TPU-v3 cores, such as selecting large batch optimizers and learning rate schedules as well as utilizing distributed evaluation and batch normalization techniques. Additionally, we present timing and performance benchmarks for EfficientNet models trained on the ImageNet dataset in order to analyze the behavior of EfficientNets at scale. With our optimizations, we are able to train EfficientNet on ImageNet to an accuracy of 83% in 1 hour and 4 minutes.

cs.LG↗

arXiv2020

The Limit of the Batch Size

Large-batch training is an efficient approach for current distributed deep learning systems. It has enabled researchers to reduce the ImageNet/ResNet-50 training from 29 hours to around 1 minute. In this paper, we focus on studying the limit of the batch size. We think it may provide a guidance to AI supercomputer and algorithm designers. We provide detailed numerical optimization instructions for step-by-step comparison. Moreover, it is important to understand the generalization and optimization performance of huge batch training. Hoffer et al. introduced "ultra-slow diffusion" theory to large-batch training. However, our experiments show contradictory results with the conclusion of Hoffer et al. We provide comprehensive experimental results and detailed analysis to study the limitations of batch size scaling and "ultra-slow diffusion" theory. For the first time we scale the batch size on ImageNet to at least a magnitude larger than all previous work, and provide detailed studies on the performance of many state-of-the-art optimization schemes under this setting. We propose an optimization recipe that is able to improve the top-1 test accuracy by 18% compared to the baseline.

cs.LG↗

arXiv2020

Communication-Optimal Tilings for Projective Nested Loops with Arbitrary Bounds

Reducing communication - either between levels of a memory hierarchy or between processors over a network - is a key component of performance optimization (in both time and energy) for many problems, including dense linear algebra, particle interactions, and machine learning. For these problems, which can be represented as nested-loop computations, previous tiling based approaches have been used to find both lower bounds on the communication required to execute them and optimal rearrangements, or blockings, to attain such lower bounds. However, such general approaches have typically assumed the problem sizes are large, an assumption that is often not met in practice. For instance, the classical $(\text{# arithmetic operations})/(\text{cache size})^{1/2}$ lower bound for matrix multiplication is not tight for matrix-vector multiplications, which must read in at least $O(\text{# arithmetic operations})$ words of memory; similar issues occur for almost all convolutions in machine learning applications, which use extremely small filter sizes (and therefore, loop bounds). In this paper, we provide an efficient way to both find and obtain, via an appropriate, efficiently constructible blocking, communication lower bounds and matching tilings which attain these lower bounds for nested loop programs with arbitrary loop bounds that operate on multidimensional arrays in the projective case, where the array indices are subsets of the loop indices. Our approach works on all such problems, regardless of dimensionality, size, memory access patterns, or number of arrays, and directly applies to (among other examples) matrix multiplication and similar dense linear algebra operations, tensor contractions, n-body pairwise interactions, pointwise convolutions, and fully connected layers.

cs.DS↗

arXiv2020

Large Batch Optimization for Deep Learning: Training BERT in 76 minutes

Training large deep neural networks on massive datasets is computationally very challenging. There has been recent surge in interest in using large batch stochastic optimization methods to tackle this issue. The most prominent algorithm in this line of research is LARS, which by employing layerwise adaptive learning rates trains ResNet on ImageNet in a few minutes. However, LARS performs poorly for attention models like BERT, indicating that its performance gains are not consistent across tasks. In this paper, we first study a principled layerwise adaptation strategy to accelerate training of deep neural networks using large mini-batches. Using this strategy, we develop a new layerwise adaptive large batch optimization technique called LAMB; we then provide convergence analysis of LAMB as well as LARS, showing convergence to a stationary point in general nonconvex settings. Our empirical results demonstrate the superior performance of LAMB across various tasks such as BERT and ResNet-50 training with very little hyperparameter tuning. In particular, for BERT training, our optimizer enables use of very large batch sizes of 32868 without any degradation of performance. By increasing the batch size to the memory limit of a TPUv3 Pod, BERT training time can be reduced from 3 days to just 76 minutes (Table 1). The LAMB implementation is available at https://github.com/tensorflow/addons/blob/master/tensorflow_addons/optimizers/lamb.py

cs.LG↗

arXiv2019

An improved analysis and unified perspective on deterministic and randomized low rank matrix approximations

We introduce a Generalized LU-Factorization (\textbf{GLU}) for low-rank matrix approximation. We relate this to past approaches and extensively analyze its approximation properties. The established deterministic guarantees are combined with sketching ensembles satisfying Johnson-Lindenstrauss properties to present complete bounds. Particularly good performance is shown for the sub-sampled randomized Hadamard transform (SRHT) ensemble. Moreover, the factorization is shown to unify and generalize many past algorithms. It also helps to explain the effect of sketching on the growth factor during Gaussian Elimination.

math.NA↗

arXiv2019

A Generalized Randomized Rank-Revealing Factorization

We introduce a Generalized Randomized QR-decomposition that may be applied to arbitrary products of matrices and their inverses, without needing to explicitly compute the products or inverses. This factorization is a critical part of a communication-optimal spectral divide-and-conquer algorithm for the nonsymmetric eigenvalue problem. In this paper, we establish that this randomized QR-factorization satisfies the strong rank-revealing properties. We also formally prove its stability, making it suitable in applications. Finally, we present numerical experiments which demonstrate that our theoretical bounds capture the empirical behavior of the factorization.

math.NA↗

arXiv2019

Large-Batch Training for LSTM and Beyond

Large-batch training approaches have enabled researchers to utilize large-scale distributed processing and greatly accelerate deep-neural net (DNN) training. For example, by scaling the batch size from 256 to 32K, researchers have been able to reduce the training time of ResNet50 on ImageNet from 29 hours to 2.2 minutes (Ying et al., 2018). In this paper, we propose a new approach called linear-epoch gradual-warmup (LEGW) for better large-batch training. With LEGW, we are able to conduct large-batch training for both CNNs and RNNs with the Sqrt Scaling scheme. LEGW enables Sqrt Scaling scheme to be useful in practice and as a result we achieve much better results than the Linear Scaling learning rate scheme. For LSTM applications, we are able to scale the batch size by a factor of 64 without losing accuracy and without tuning the hyper-parameters. For CNN applications, LEGW is able to achieve the same accuracy even as we scale the batch size to 32K. LEGW works better than previous large-batch auto-tuning techniques. LEGW achieves a 5.3X average speedup over the baselines for four LSTM-based applications on the same hardware. We also provide some theoretical explanations for LEGW.

cs.LG↗