We propose two coded schemes for the distributed computing problem of multiplying a matrix by a set of vectors. The first scheme is based on partitioning the matrix into submatrices and applying maximum distance separable (MDS) codes to each submatrix. For this scheme, we prove that up to a given number of partitions the communication load and the computational delay (not including the encoding and decoding delay) are identical to those of the scheme recently proposed by Li et al., based on a single, long MDS code. However, due to the use of shorter MDS codes, our scheme yields a significantly lower overall computational delay when the delay incurred by encoding and decoding is also considered. We further propose a second coded scheme based on Luby Transform (LT) codes under inactivation decoding. Interestingly, LT codes may reduce the delay over the partitioned scheme at the expense of an increased communication load. We also consider distributed computing under a deadline and show numerically that the proposed schemes outperform other schemes in the literature, with the LT code-based scheme yielding the best performance for the scenarios considered.
Modern learning models are characterized by large hyperparameter spaces and long training times. These properties, coupled with the rise of parallel computing and the growing demand to productionize machine learning workloads, motivate the need to develop mature hyperparameter optimization functionality in distributed computing settings. We address this challenge by first introducing a simple and robust hyperparameter optimization algorithm called ASHA, which exploits parallelism and aggressive early-stopping to tackle large-scale hyperparameter optimization problems. Our extensive empirical results show that ASHA outperforms existing state-of-the-art hyperparameter optimization methods; scales linearly with the number of workers in distributed settings; and is suitable for massive parallelism, as demonstrated on a task with 500 workers. We then describe several design decisions we encountered, along with our associated solutions, when integrating ASHA in Determined AI's end-to-end production-quality machine learning system that offers hyperparameter tuning as a service.
We propose a coded distributed computing scheme based on Raptor codes to address the straggler problem. In particular, we consider a scheme where each server computes intermediate values, referred to as droplets, that are either stored locally or sent over the network. Once enough droplets are collected, the computation can be completed. Compared to previous schemes in the literature, our proposed scheme achieves lower computational delay when the decoding time is taken into account.
Humans learn complex latent structures from their environments (e.g., natural language, mathematics, music, social hierarchies). In cognitive science and cognitive neuroscience, models that infer higher-order structures from sensory or first-order representations have been proposed to account for the complexity and flexibility of human behavior. But how do the structures that these models invoke arise in neural systems in the first place? To answer this question, we explain how a system can learn latent representational structures (i.e., predicates) from experience with wholly unstructured data. During the process of predicate learning, an artificial neural network exploits the naturally occurring dynamic properties of distributed computing across neuronal assemblies in order to learn predicates, but also to combine them compositionally, two computational aspects which appear to be necessary for human behavior as per formal theories in multiple domains. We describe how predicates can be combined generatively using neural oscillations to achieve human-like extrapolation and compositionality in an artificial neural network. The ability to learn predicates from experience, to represent structures compositionally, and to extrapolate to unseen data offers an inroads to understanding and modeling the most complex human behaviors.
In the United States, heart disease is the leading cause of death for both men and women, accounting for 610,000 deaths each year [1]. Physicians use Magnetic Resonance Imaging (MRI) scans to take images of the heart in order to non-invasively estimate its structural and functional parameters for cardiovascular diagnosis and disease management. The end-systolic volume (ESV) and end-diastolic volume (EDV) of the left ventricle (LV), and the ejection fraction (EF) are indicators of heart disease. These measures can be derived from the segmented contours of the LV; thus, consistent and accurate segmentation of the LV from MRI images are critical to the accuracy of the ESV, EDV, and EF, and to non-invasive cardiac disease detection. In this work, various image preprocessing techniques, model configurations using the U-Net deep learning architecture, postprocessing methods, and approaches for volume estimation are investigated. An end-to-end analytics pipeline with multiple stages is provided for automated LV segmentation and volume estimation. First, image data are reformatted and processed from DICOM and NIfTI formats to raw images in array format. Secondly, raw images are processed with multiple image preprocessing methods and cropped to include only the Region of Interest (ROI). Thirdly, preprocessed images are segmented using U-Net models. Lastly, post processing of segmented images to remove extra contours along with intelligent slice and frame selection are applied, followed by calculation of the ESV, EDV, and EF. This analytics pipeline is implemented and runs on a distributed computing environment with a GPU cluster at the San Diego Supercomputer Center at UCSD.
PARyOpt is a python based implementation of the Bayesian optimization routine designed for remote and asynchronous function evaluations. Bayesian optimization is especially attractive for computational optimization due to its low cost function footprint as well as the ability to account for uncertainties in data. A key challenge to efficiently deploy any optimization strategy on distributed computing systems is the synchronization step, where data from multiple function calls is assimilated to identify the next campaign of function calls. Bayesian optimization provides an elegant approach to overcome this issue via asynchronous updates. We formulate, develop and implement a parallel, asynchronous variant of Bayesian optimization. The framework is robust and resilient to external failures. We show how such asynchronous evaluations help reduce the total optimization wall clock time for a suite of test problems. Additionally, we show how the software design of the framework allows easy extension to response surface reconstruction (Kriging), providing a high performance software for autonomous exploration. The software is available on PyPI, with examples and documentation.
We propose a framework stitching of vector representations of large scale raster mosaic images in distributed computing model. In this way, the negative effect of the lack of resources of the central system and scalability problem can be eliminated. The product obtained by this study can be used in applications requiring spatial and temporal analysis on big satellite map images. This study also shows that big data frameworks are not only used in applications of text-based data mining and machine learning algorithms, but also used in applications of algorithms in image processing. The effectiveness of the product realized with this project is also going to be proven by scalability and performance tests performed on real world LandSat-8 satellite images.
We study the expected completion time of some recently proposed algorithms for distributed computing which redundantly assign computing tasks to multiple machines in order to tolerate a certain number of machine failures. We analytically show that not only the amount of redundancy but also the task-to-machine assignments affect the latency in a distributed system. We study systems with a fixed number of computing tasks that are split in possibly overlapping batches, and independent exponentially distributed machine service times. We show that, for such systems, the uniform replication of non- overlapping (disjoint) batches of computing tasks achieves the minimum expected computing time.
State-of-the-art link prediction utilizes combinations of complex features derived from network panel data. We here show that computationally less expensive features can achieve the same performance in the common scenario in which the data is available as a sequence of interactions. Our features are based on social vector clocks, an adaptation of the vector-clock concept introduced in distributed computing to social interaction networks. In fact, our experiments suggest that by taking into account the order and spacing of interactions, social vector clocks exploit different aspects of link formation so that their combination with previous approaches yields the most accurate predictor to date.
Analysis of the convergence rates of modern convex optimization algorithms can be achived through binary means: analysis of emperical convergence, or analysis of theoretical convergence. These two pathways of capturing information diverge in efficacy when moving to the world of distributed computing, due to the introduction of non-intuitive, non-linear slowdowns associated with broadcasting, and in some cases, gathering operations. Despite these nuances in the rates of convergence, we can still show the existence of convergence, and lower bounds for the rates. This paper will serve as a helpful cheat-sheet for machine learning practitioners encountering this problem class in the field.
The cellular tree classifier model addresses a fundamental problem in the design of classifiers for a parallel or distributed computing world: Given a data set, is it sufficient to apply a majority rule for classification, or shall one split the data into two or more parts and send each part to a potentially different computer (or cell) for further processing? At first sight, it seems impossible to define with this paradigm a consistent classifier as no cell knows the "original data size", $n$. However, we show that this is not so by exhibiting two different consistent classifiers. The consistency is universal but is only shown for distributions with nonatomic marginals.
This paper proposes a method to estimate the total time required to solve SAT in distributed environments via partitioning approach. It is based on the observation that for some simple forms of problem partitioning one can use the Monte Carlo approach to estimate the time required to solve an original problem. The method proposed is based on an algorithm for searching for partitioning with an optimal solving time estimation. We applied this method to estimate the time required to perform logical cryptanalysis of the widely known stream ciphers A5/1 and Bivium. The paper also describes a volunteer computing project SAT@home aimed at solving hard combinatorial problems reduced to SAT. In this project during several months there were solved 10 problems of logical cryptanalysis of the A5/1 cipher thatcould not be solved using known rainbow tables.
If learning methods are to scale to the massive sizes of modern datasets, it is essential for the field of machine learning to embrace parallel and distributed computing. Inspired by the recent development of matrix factorization methods with rich theory but poor computational complexity and by the relative ease of mapping matrices onto distributed architectures, we introduce a scalable divide-and-conquer framework for noisy matrix factorization. We present a thorough theoretical analysis of this framework in which we characterize the statistical errors introduced by the "divide" step and control their magnitude in the "conquer" step, so that the overall algorithm enjoys high-probability estimation guarantees comparable to those of its base algorithm. We also present experiments in collaborative filtering and video background modeling that demonstrate the near-linear to superlinear speed-ups attainable with this approach.
Training large machine learning (ML) models with many variables or parameters can take a long time if one employs sequential procedures even with stochastic updates. A natural solution is to turn to distributed computing on a cluster; however, naive, unstructured parallelization of ML algorithms does not usually lead to a proportional speedup and can even result in divergence, because dependencies between model elements can attenuate the computational gains from parallelization and compromise correctness of inference. Recent efforts toward this issue have benefited from exploiting the static, a priori block structures residing in ML algorithms. In this paper, we take this path further by exploring the dynamic block structures and workloads therein present during ML program execution, which offers new opportunities for improving convergence, correctness, and load balancing in distributed ML. We propose and showcase a general-purpose scheduler, STRADS, for coordinating distributed updates in ML algorithms, which harnesses the aforementioned opportunities in a systematic way. We provide theoretical guarantees for our scheduler, and demonstrate its efficacy versus static block structures on Lasso and Matrix Factorization.
In the last two decades, modal and description logics have been applied to numerous areas of computer science, including knowledge representation, formal verification, database theory, distributed computing and, more recently, semantic web and ontologies. For this reason, the problem of automated reasoning in modal and description logics has been thoroughly investigated. In particular, many approaches have been proposed for efficiently handling the satisfiability of the core normal modal logic K(m), and of its notational variant, the description logic ALC. Although simple in structure, K(m)/ALC is computationally very hard to reason on, its satisfiability being PSPACE-complete. In this paper we start exploring the idea of performing automated reasoning tasks in modal and description logics by encoding them into SAT, so that to be handled by state-of-the-art SAT tools; as with most previous approaches, we begin our investigation from the satisfiability in K(m). We propose an efficient encoding, and we test it on an extensive set of benchmarks, comparing the approach with the main state-of-the-art tools available. Although the encoding is necessarily worst-case exponential, from our experiments we notice that, in practice, this approach can handle most or all the problems which are at the reach of the other approaches, with performances which are comparable with, or even better than, those of the current state-of-the-art tools.
The kernel $k$-means is an effective method for data clustering which extends the commonly-used $k$-means algorithm to work on a similarity matrix over complex data structures. The kernel $k$-means algorithm is however computationally very complex as it requires the complete data matrix to be calculated and stored. Further, the kernelized nature of the kernel $k$-means algorithm hinders the parallelization of its computations on modern infrastructures for distributed computing. In this paper, we are defining a family of kernel-based low-dimensional embeddings that allows for scaling kernel $k$-means on MapReduce via an efficient and unified parallelization strategy. Afterwards, we propose two methods for low-dimensional embedding that adhere to our definition of the embedding family. Exploiting the proposed parallelization strategy, we present two scalable MapReduce algorithms for kernel $k$-means. We demonstrate the effectiveness and efficiency of the proposed algorithms through an empirical evaluation on benchmark data sets.
Computational Grids are a new trend in distributed computing systems. They allow the sharing of geographically distributed resources in an efficient way, extending the boundaries of what we perceive as distributed computing. Various sciences can benefit from the use of grids to solve CPU-intensive problems, creating potential benefits to the entire society. Job scheduling is an integrated part of parallel and distributed computing. It allows selecting correct match of resource for a particular job and thus increases the job throughput and utilization of resources. Job should be scheduled in an automatic way to make the system more reliable, accessible and less sensitive to subsystem failures. This paper provides a survey on various heuristic algorithms, used for scheduling in grid.
We present chemlambda (or the chemical concrete machine), an artificial chemistry with the following properties: (a) is Turing complete, (b) has a model of decentralized, distributed computing associated to it, (c) works at the level of individual (artificial) molecules, subject of reversible, but otherwise deterministic interactions with a small number of enzymes, (d) encodes information in the geometrical structure of the molecules and not in their numbers, (e) all interactions are purely local in space and time. This is part of a larger project to create computing, artificial chemistry and artificial life in a distributed context, using topological and graphical languages.
We present a machine learning framework for modeling protein dynamics. Our approach uses L1-regularized, reversible hidden Markov models to understand large protein datasets generated via molecular dynamics simulations. Our model is motivated by three design principles: (1) the requirement of massive scalability; (2) the need to adhere to relevant physical law; and (3) the necessity of providing accessible interpretations, critical for both cellular biology and rational drug design. We present an EM algorithm for learning and introduce a model selection criteria based on the physical notion of convergence in relaxation timescales. We contrast our model with standard methods in biophysics and demonstrate improved robustness. We implement our algorithm on GPUs and apply the method to two large protein simulation datasets generated respectively on the NCSA Bluewaters supercomputer and the Folding@Home distributed computing network. Our analysis identifies the conformational dynamics of the ubiquitin protein critical to cellular signaling, and elucidates the stepwise activation mechanism of the c-Src kinase protein.
Many problems in machine learning and other fields can be (re)for-mulated as linearly constrained separable convex programs. In most of the cases, there are multiple blocks of variables. However, the traditional alternating direction method (ADM) and its linearized version (LADM, obtained by linearizing the quadratic penalty term) are for the two-block case and cannot be naively generalized to solve the multi-block case. So there is great demand on extending the ADM based methods for the multi-block case. In this paper, we propose LADM with parallel splitting and adaptive penalty (LADMPSAP) to solve multi-block separable convex programs efficiently. When all the component objective functions have bounded subgradients, we obtain convergence results that are stronger than those of ADM and LADM, e.g., allowing the penalty parameter to be unbounded and proving the sufficient and necessary conditions} for global convergence. We further propose a simple optimality measure and reveal the convergence rate of LADMPSAP in an ergodic sense. For programs with extra convex set constraints, with refined parameter estimation we devise a practical version of LADMPSAP for faster convergence. Finally, we generalize LADMPSAP to handle programs with more difficult objective functions by linearizing part of the objective function as well. LADMPSAP is particularly suitable for sparse representation and low-rank recovery problems because its subproblems have closed form solutions and the sparsity and low-rankness of the iterates can be preserved during the iteration. It is also highly parallelizable and hence fits for parallel or distributed computing. Numerical experiments testify to the advantages of LADMPSAP in speed and numerical accuracy.