This article introduces new multiplicative updates for nonnegative matrix factorization with the $\beta$-divergence and sparse regularization of one of the two factors (say, the activation matrix). It is well known that the norm of the other factor (the dictionary matrix) needs to be controlled in order to avoid an ill-posed formulation. Standard practice consists in constraining the columns of the dictionary to have unit norm, which leads to a nontrivial optimization problem. Our approach leverages a reparametrization of the original problem into the optimization of an equivalent scale-invariant objective function. From there, we derive block-descent majorization-minimization algorithms that result in simple multiplicative updates for either $\ell_{1}$-regularization or the more "aggressive" log-regularization. In contrast with other state-of-the-art methods, our algorithms are universal in the sense that they can be applied to any $\beta$-divergence (i.e., any value of $\beta$) and that they come with convergence guarantees. We report numerical comparisons with existing heuristic and Lagrangian methods using various datasets: face images, an audio spectrogram, hyperspectral data, and song play counts. We show that our methods obtain solutions of similar quality at convergence (similar objective values) but with significantly reduced CPU times.
K-nearest neighbor search is one of the fundamental tasks in various applications and the hierarchical navigable small world (HNSW) has recently drawn attention in large-scale cloud services, as it easily scales up the database while offering fast search. On the other hand, a computational storage device (CSD) that combines programmable logic and storage modules on a single board becomes popular to address the data bandwidth bottleneck of modern computing systems. In this paper, we propose a computational storage platform that can accelerate a large-scale graph-based nearest neighbor search algorithm based on SmartSSD CSD. To this end, we modify the algorithm more amenable on the hardware and implement two types of accelerators using HLS- and RTL-based methodology with various optimization methods. In addition, we scale up the proposed platform to have 4 SmartSSDs and apply graph parallelism to boost the system performance further. As a result, the proposed computational storage platform achieves 75.59 query per second throughput for the SIFT1B dataset at 258.66W power dissipation, which is 12.83x and 17.91x faster and 10.43x and 24.33x more energy efficient than the conventional CPU-based and GPU-based server platform, respectively. With multi-terabyte storage and custom acceleration capability, we believe that the proposed computational storage platform is a promising solution for cost-sensitive cloud datacenters.
Natural Language Processing (NLP) inference is seeing increasing adoption by mobile applications, where on-device inference is desirable for crucially preserving user data privacy and avoiding network roundtrips. Yet, the unprecedented size of an NLP model stresses both latency and memory, creating a tension between the two key resources of a mobile device. To meet a target latency, holding the whole model in memory launches execution as soon as possible but increases one app's memory footprints by several times, limiting its benefits to only a few inferences before being recycled by mobile memory management. On the other hand, loading the model from storage on demand incurs IO as long as a few seconds, far exceeding the delay range satisfying to a user; pipelining layerwise model loading and execution does not hide IO either, due to the high skewness between IO and computation delays. To this end, we propose Speedy Transformer Inference (STI). Built on the key idea of maximizing IO/compute resource utilization on the most important parts of a model, STI reconciles the latency v.s. memory tension via two novel techniques. First, model sharding. STI manages model parameters as independently tunable shards, and profiles their importance to accuracy. Second, elastic pipeline planning with a preload buffer. STI instantiates an IO/compute pipeline and uses a small buffer for preload shards to bootstrap execution without stalling at early stages; it judiciously selects, tunes, and assembles shards per their importance for resource-elastic execution, maximizing inference accuracy. Atop two commodity SoCs, we build STI and evaluate it against a wide range of NLP tasks, under a practical range of target latencies, and on both CPU and GPU. We demonstrate that STI delivers high accuracies with 1-2 orders of magnitude lower memory, outperforming competitive baselines.
BLAST is a standard tool in bioinformatics for creating local sequence alignments using a "seed-and-extend" approach. Here we introduce an analogous seed-and-extend algorithm that produces local network alignments: BLANT (Basic Local Alignment of Network Topology). In Part 1, we introduced BLANT-seed, which generates graphlet-based seeds using only topological information. Here, in Part 2, we describe BLANT-extend, which "grows" seeds to larger local alignments using only topological information. We allow the user to specify bounds on several measures an alignment must satisfy, including the edge density, edge commonality (i.e., aligned edges), and node-pair similarity if such a measure is used; the latter allows the inclusion of sequence-based similarity, if desired, as well as local topological constraints. BLANT-extend is able to enumerate all possible alignments satisfying the bounds that can be grown from each seed, within a specified CPU time or number of generated alignments. While topology-driven local network alignment has a wide variety of potential applications outside bioinformatics, here we focus on the alignment of Protein-Protein Interaction (PPI) networks. We show that BLANT is capable of finding large, high-quality local alignments when the networks are known to have high topological similarity -- for example recovering hundreds of orthologs between networks of the recent Integrated Interaction Database (IID). Predictably, however, it performs less well when true topological similarity is absent, as is the case in most current experimental PPI networks that are noisy and have wide disparity in edge density which results in low common coverage.
In this study, we present a deep learning-optimization framework to tackle dynamic mixed-integer programs. Specifically, we develop a bidirectional Long Short Term Memory (LSTM) framework that can process information forward and backward in time to learn optimal solutions to sequential decision-making problems. We demonstrate our approach in predicting the optimal decisions for the single-item capacitated lot-sizing problem (CLSP), where a binary variable denotes whether to produce in a period or not. Due to the dynamic nature of the problem, the CLSP can be treated as a sequence labeling task where a recurrent neural network can capture the problem's temporal dynamics. Computational results show that our LSTM-Optimization (LSTM-Opt) framework significantly reduces the solution time of benchmark CLSP problems without much loss in feasibility and optimality. For example, the predictions at the 85\% level reduce the CPLEX solution time by a factor of 9 on average for over 240,000 test instances with an optimality gap of less than 0.05\% and 0.4\% infeasibility in the test set. Also, models trained using shorter planning horizons can successfully predict the optimal solution of the instances with longer planning horizons. For the hardest data set, the LSTM predictions at the 25\% level reduce the solution time of 70 CPU hours to less than 2 CPU minutes with an optimality gap of 0.8\% and without any infeasibility. The LSTM-Opt framework outperforms classical ML algorithms, such as the logistic regression and random forest, in terms of the solution quality, and exact approaches, such as the ($\ell$, S) and dynamic programming-based inequalities, with respect to the solution time improvement. Our machine learning approach could be beneficial in tackling sequential decision-making problems similar to CLSP, which need to be solved repetitively, frequently, and in a fast manner.
Artificial Intelligence for IT Operations (AIOps) describes the process of maintaining and operating large IT systems using diverse AI-enabled methods and tools for, e.g., anomaly detection and root cause analysis, to support the remediation, optimization, and automatic initiation of self-stabilizing IT activities. The core step of any AIOps workflow is anomaly detection, typically performed on high-volume heterogeneous data such as log messages (logs), metrics (e.g., CPU utilization), and distributed traces. In this paper, we propose a method for reliable and practical anomaly detection from system logs. It overcomes the common disadvantage of related works, i.e., the need for a large amount of manually labeled training data, by building an anomaly detection model with log instructions from the source code of 1000+ GitHub projects. The instructions from diverse systems contain rich and heterogenous information about many different normal and abnormal IT events and serve as a foundation for anomaly detection. The proposed method, named ADLILog, combines the log instructions and the data from the system of interest (target system) to learn a deep neural network model through a two-phase learning procedure. The experimental results show that ADLILog outperforms the related approaches by up to 60% on the F1 score while satisfying core non-functional requirements for industrial deployments such as unsupervised design, efficient model updates, and small model sizes.
The accelerating technological landscape and drive towards net-zero emission made the power system grow in scale and complexity. Serial computational approaches for grid planning and operation struggle to execute necessary calculations within reasonable times. Resorting to high-performance and parallel computing approaches has become paramount. Moreover, the ambitious plans for the future grid and IoT integration make a shift towards utilizing Cloud computing inevitable. This article recounts the dawn of parallel computation and its appearance in power system studies, reviewing the most recent literature and research on exploiting the available computational resources and technologies today. The article starts with a brief introduction to the field. The relevant hardware and paradigms are then explained thoroughly in this article providing a base for the reader to understand the literature. Later, parallel power system studies are reviewed, reciting the study development from older papers up to the 21st century, emphasizing the most impactful work of the last decade. The studies included system stability studies, state estimation and power system operation, and market optimization. The reviews are split into \ac{CPU} based,\ac{GPU} based, and Cloud-based studies. Finally, the state-of-the-art is discussed, highlighting the issue of standardization and the future of computation in power system studies.
The past several years have witnessed the success of transformer-based models, and their scale and application scenarios continue to grow aggressively. The current landscape of transformer models is increasingly diverse: the model size varies drastically with the largest being of hundred-billion parameters; the model characteristics differ due to the sparsity introduced by the Mixture-of-Experts; the target application scenarios can be latency-critical or throughput-oriented; the deployment hardware could be single- or multi-GPU systems with different types of memory and storage, etc. With such increasing diversity and the fast-evolving pace of transformer models, designing a highly performant and efficient inference system is extremely challenging. In this paper, we present DeepSpeed Inference, a comprehensive system solution for transformer model inference to address the above-mentioned challenges. DeepSpeed Inference consists of (1) a multi-GPU inference solution to minimize latency while maximizing the throughput of both dense and sparse transformer models when they fit in aggregate GPU memory, and (2) a heterogeneous inference solution that leverages CPU and NVMe memory in addition to the GPU memory and compute to enable high inference throughput with large models which do not fit in aggregate GPU memory. DeepSpeed Inference reduces latency by up to 7.3X over the state-of-the-art for latency-oriented scenarios and increases throughput by over 1.5x for throughput-oriented scenarios. Moreover, it enables trillion parameter scale inference under real-time latency constraints by leveraging hundreds of GPUs, an unprecedented scale for inference. It can inference 25x larger models than with GPU-only solutions, while delivering a high throughput of 84 TFLOPS (over $50\%$ of A6000 peak).
The attention mechanisms of transformers effectively extract pertinent information from the input sequence. However, the quadratic complexity of self-attention w.r.t the sequence length incurs heavy computational and memory burdens, especially for tasks with long sequences. Existing accelerators face performance degradation in these tasks. To this end, we propose SALO to enable hybrid sparse attention mechanisms for long sequences. SALO contains a data scheduler to map hybrid sparse attention patterns onto hardware and a spatial accelerator to perform the efficient attention computation. We show that SALO achieves 17.66x and 89.33x speedup on average compared to GPU and CPU implementations, respectively, on typical workloads, i.e., Longformer and ViL.
Constrained partially observable Markov decision processes (CPOMDPs) have been used to model various real-world phenomena. However, they are notoriously difficult to solve to optimality, and there exist only a few approximation methods for obtaining high-quality solutions. In this study, grid-based approximations are used in combination with linear programming (LP) models to generate approximate policies for CPOMDPs. A detailed numerical study is conducted with six CPOMDP problem instances considering both their finite and infinite horizon formulations. The quality of approximation algorithms for solving unconstrained POMDP problems is established through a comparative analysis with exact solution methods. Then, the performance of the LP-based CPOMDP solution approaches for varying budget levels is evaluated. Finally, the flexibility of LP-based approaches is demonstrated by applying deterministic policy constraints, and a detailed investigation into their impact on rewards and CPU run time is provided. For most of the finite horizon problems, deterministic policy constraints are found to have little impact on expected reward, but they introduce a significant increase to CPU run time. For infinite horizon problems, the reverse is observed: deterministic policies tend to yield lower expected total rewards than their stochastic counterparts, but the impact of deterministic constraints on CPU run time is negligible in this case. Overall, these results demonstrate that LP models can effectively generate approximate policies for both finite and infinite horizon problems while providing the flexibility to incorporate various additional constraints into the underlying model.
Improving software performance is an important yet challenging part of the software development cycle. Today, the majority of performance inefficiencies are identified and patched by performance experts. Recent advancements in deep learning approaches and the wide-spread availability of open source data creates a great opportunity to automate the identification and patching of performance problems. In this paper, we present DeepPERF, a transformer-based approach to suggest performance improvements for C# applications. We pretrain DeepPERF on English and Source code corpora and followed by finetuning for the task of generating performance improvement patches for C# applications. Our evaluation shows that our model can generate the same performance improvement suggestion as the developer fix in ~53% of the cases, getting ~34% of them verbatim in our expert-verified dataset of performance changes made by C# developers. Additionally, we evaluate DeepPERF on 50 open source C# repositories on GitHub using both benchmark and unit tests and find that our model is able to suggest valid performance improvements that can improve both CPU usage and Memory allocations. So far we've submitted 19 pull-requests with 28 different performance optimizations and 11 of these PRs have been approved by the project owners.
Gaussian Processes (GPs) are highly expressive, probabilistic models. A major limitation is their computational complexity. Naively, exact GP inference requires $\mathcal{O}(N^3)$ computations with $N$ denoting the number of modeled points. Current approaches to overcome this limitation either rely on sparse, structured or stochastic representations of data or kernel respectively and usually involve nested optimizations to evaluate a GP. We present a new, generative method named Iterative Charted Refinement (ICR) to model GPs on nearly arbitrarily spaced points in $\mathcal{O}(N)$ time for decaying kernels without nested optimizations. ICR represents long- as well as short-range correlations by combining views of the modeled locations at varying resolutions with a user-provided coordinate chart. In our experiment with points whose spacings vary over two orders of magnitude, ICR's accuracy is comparable to state-of-the-art GP methods. ICR outperforms existing methods in terms of computational speed by one order of magnitude on the CPU and GPU and has already been successfully applied to model a GP with $122$ billion parameters.
Machine learning (ML) has entered the mobile era where an enormous number of ML models are deployed on edge devices. However, running common ML models on edge devices continuously may generate excessive heat from the computation, forcing the device to "slow down" to prevent overheating, a phenomenon called thermal throttling. This paper studies the impact of thermal throttling on mobile phones: when it occurs, the CPU clock frequency is reduced, and the model inference latency may increase dramatically. This unpleasant inconsistent behavior has a substantial negative effect on user experience, but it has been overlooked for a long time. To counter thermal throttling, we propose to utilize dynamic networks with shared weights and dynamically shift between large and small ML models seamlessly according to their thermal profile, i.e., shifting to a small model when the system is about to throttle. With the proposed dynamic shifting, the application runs consistently without experiencing CPU clock frequency degradation and latency increase. In addition, we also study the resulting accuracy when dynamic shifting is deployed and show that our approach provides a reasonable trade-off between model latency and model accuracy.
As Deep Neural Networks (DNNs) have become an increasingly ubiquitous workload, the range of libraries and tooling available to aid in their development and deployment has grown significantly. Scalable, production quality tools are freely available under permissive licenses, and are accessible enough to enable even small teams to be very productive. However within the research community, awareness and usage of said tools is not necessarily widespread, and researchers may be missing out on potential productivity gains from exploiting the latest tools and workflows. This paper presents a case study where we discuss our recent experience producing an end-to-end artificial intelligence application for industrial defect detection. We detail the high level deep learning libraries, containerized workflows, continuous integration/deployment pipelines, and open source code templates we leveraged to produce a competitive result, matching the performance of other ranked solutions to our three target datasets. We highlight the value that exploiting such systems can bring, even for research, and detail our solution and present our best results in terms of accuracy and inference time on a server class GPU, as well as inference times on a server class CPU, and a Raspberry Pi 4.
With the yearning for deep learning democratization, there are increasing demands to implement Transformer-based natural language processing (NLP) models on resource-constrained devices for low-latency and high accuracy. Existing BERT pruning methods require domain experts to heuristically handcraft hyperparameters to strike a balance among model size, latency, and accuracy. In this work, we propose AE-BERT, an automatic and efficient BERT pruning framework with efficient evaluation to select a "good" sub-network candidate (with high accuracy) given the overall pruning ratio constraints. Our proposed method requires no human experts experience and achieves a better accuracy performance on many NLP tasks. Our experimental results on General Language Understanding Evaluation (GLUE) benchmark show that AE-BERT outperforms the state-of-the-art (SOTA) hand-crafted pruning methods on BERT$_{\mathrm{BASE}}$. On QNLI and RTE, we obtain 75\% and 42.8\% more overall pruning ratio while achieving higher accuracy. On MRPC, we obtain a 4.6 higher score than the SOTA at the same overall pruning ratio of 0.5. On STS-B, we can achieve a 40\% higher pruning ratio with a very small loss in Spearman correlation compared to SOTA hand-crafted pruning methods. Experimental results also show that after model compression, the inference time of a single BERT$_{\mathrm{BASE}}$ encoder on Xilinx Alveo U200 FPGA board has a 1.83$\times$ speedup compared to Intel(R) Xeon(R) Gold 5218 (2.30GHz) CPU, which shows the reasonableness of deploying the proposed method generated subnets of BERT$_{\mathrm{BASE}}$ model on computation restricted devices.
GPUs have been favored for training deep learning models due to their highly parallelized architecture. As a result, most studies on training optimization focus on GPUs. There is often a trade-off, however, between cost and efficiency when deciding on how to choose the proper hardware for training. In particular, CPU servers can be beneficial if training on CPUs was more efficient, as they incur fewer hardware update costs and better utilizing existing infrastructure. This paper makes several contributions to research on training deep learning models using CPUs. First, it presents a method for optimizing the training of deep learning models on Intel CPUs and a toolkit called ProfileDNN, which we developed to improve performance profiling. Second, we describe a generic training optimization method that guides our workflow and explores several case studies where we identified performance issues and then optimized the Intel Extension for PyTorch, resulting in an overall 2x training performance increase for the RetinaNet-ResNext50 model. Third, we show how to leverage the visualization capabilities of ProfileDNN, which enabled us to pinpoint bottlenecks and create a custom focal loss kernel that was two times faster than the official reference PyTorch implementation.
Recently, GAN-based neural vocoders such as Parallel WaveGAN, MelGAN, HiFiGAN, and UnivNet have become popular due to their lightweight and parallel structure, resulting in a real-time synthesized waveform with high fidelity, even on a CPU. HiFiGAN and UnivNet are two SOTA vocoders. Despite their high quality, there is still room for improvement. In this paper, motivated by the structure of Vision Outlooker from computer vision, we adopt a similar idea and propose an effective and lightweight neural vocoder called WOLONet. In this network, we develop a novel lightweight block that uses a location-variable, channel-independent, and depthwise dynamic convolutional kernel with sinusoidally activated dynamic kernel weights. To demonstrate the effectiveness and generalizability of our method, we perform an ablation study to verify our novel design and make a subjective and objective comparison with typical GAN-based vocoders. The results show that our WOLONet achieves the best generation quality while requiring fewer parameters than the two neural SOTA vocoders, HiFiGAN and UnivNet.
A digital twin is a surrogate model that has the main feature to mirror the original process behavior. Associating the dynamical process with a digital twin model of reduced complexity has the significant advantage to map the dynamics with high accuracy and reduced costs in CPU time and hardware to timescales over which that suffers significantly changes and so it is difficult to explore. This paper introduces a new framework for creating efficient digital twin models of fluid flows. We introduce a novel algorithm that combines the advantages of Krylov based dynamic mode decomposition with proper orthogonal decomposition and outperforms the selection of the most influential modes. We prove that randomized orthogonal decomposition algorithm provides several advantages over SVD empirical orthogonal decomposition methods and mitigates the projection error formulating a multiobjective optimization problem.We involve the state-of-the-art artificial intelligence Deep Learning (DL) to perform a real-time adaptive calibration of the digital twin model, with increasing fidelity. The output is a high-fidelity DIGITAL TWIN DATA MODEL of the fluid flow dynamics, with the advantage of a reduced complexity. The new modelling tools are investigated in the numerical simulation of three wave phenomena with increasing complexity. We show that the outputs are consistent with the original source data.We perform a thorough assessment of the performance of the new digital twin data models, in terms of numerical accuracy and computational efficiency, including a time simulation response feature study.
Time series alignment methods call for highly expressive, differentiable and invertible warping functions which preserve temporal topology, i.e diffeomorphisms. Diffeomorphic warping functions can be generated from the integration of velocity fields governed by an ordinary differential equation (ODE). Gradient-based optimization frameworks containing diffeomorphic transformations require to calculate derivatives to the differential equation's solution with respect to the model parameters, i.e. sensitivity analysis. Unfortunately, deep learning frameworks typically lack automatic-differentiation-compatible sensitivity analysis methods; and implicit functions, such as the solution of ODE, require particular care. Current solutions appeal to adjoint sensitivity methods, ad-hoc numerical solvers or ResNet's Eulerian discretization. In this work, we present a closed-form expression for the ODE solution and its gradient under continuous piecewise-affine (CPA) velocity functions. We present a highly optimized implementation of the results on CPU and GPU. Furthermore, we conduct extensive experiments on several datasets to validate the generalization ability of our model to unseen data for time-series joint alignment. Results show significant improvements both in terms of efficiency and accuracy.
Low-Power Edge-AI capabilities are essential for on-device extended reality (XR) applications to support the vision of Metaverse. In this work, we investigate two representative XR workloads: (i) Hand detection and (ii) Eye segmentation, for hardware design space exploration. For both applications, we train deep neural networks and analyze the impact of quantization and hardware specific bottlenecks. Through simulations, we evaluate a CPU and two systolic inference accelerator implementations. Next, we compare these hardware solutions with advanced technology nodes. The impact of integrating state-of-the-art emerging non-volatile memory technology (STT/SOT/VGSOT MRAM) into the XR-AI inference pipeline is evaluated. We found that significant energy benefits (>=24%) can be achieved for hand detection (IPS=10) and eye segmentation (IPS=0.1) by introducing non-volatile memory in the memory hierarchy for designs at 7nm node while meeting minimum IPS (inference per second). Moreover, we can realize substantial reduction in area (>=30%) owing to the small form factor of MRAM compared to traditional SRAM.