Given a graph G of n nodes partitioned into facilities and customers, the r-edge interdiction covering problem (REIC) is to remove up to r edges so as to maximize the total weight of customers disconnected from all facilities, which is called the covering objective function. While REIC is known to be NP-complete for general graphs, Fröhlich and Ruzika show that the problem can be solved in polynomial time when G is a tree, providing an O(n^7 r)-time algorithm. We give an efficient O(nr^2)-time dynamic programming algorithm for REIC on trees that is fixed-parameter linear in n. Evaluating our solution on a benchmark of randomly generated tree networks with baselines of the Fröhlich and Ruzika algorithm and the Gurobi integer program solver, we demonstrate that in practice, our algorithm is both significantly faster and less sensitive to network topology and size. We extend our algorithm for REIC to graphs of bounded treewidth, a well-studied family of sparse graphs that generalizes trees, and obtain a matching runtime of O(nr^2). We also consider the r-facility interdiction covering problem (RFIC), a novel variant of this network interdiction problem where the goal is to remove up to r facilities to maximize the covering objective function over disconnected customers. We show that RFIC is NP-complete by observing it generalizes the small set bipartite vertex expansion problem (SSBVE), also known as the minimum p-union problem. We give an O(nr^2)-time algorithm for RFIC on trees, which also gives an O(n^3)-time algorithm for SSBVE on trees.

Deploying deep neural networks on mobile devices is increasingly important but remains challenging due to limited computing resources. On the other hand, their unified memory architecture and narrower gap between CPU and GPU performance provide an opportunity to reduce inference latency by assigning tasks to both CPU and GPU. The main obstacles for such collaborative execution are the significant synchronization overhead required to combine partial results, and the difficulty of predicting execution times of tasks assigned to CPU and GPU (due to the dynamic selection of implementations and parallelism level). To overcome these obstacles, we propose both a lightweight synchronization mechanism based on OpenCL fine-grained shared virtual memory (SVM) and machine learning models to accurately predict execution times. Notably, these models capture the performance characteristics of GPU kernels and account for their dispatch times. A comprehensive evaluation on four mobile platforms shows that our approach can quickly select CPU-GPU co-execution strategies achieving up to 1.89x speedup for linear layers and 1.75x speedup for convolutional layers (close to the achievable maximum values of 2.01x and 1.87x, respectively, found by exhaustive grid search on a Pixel 5 smartphone).

Given the significant advances in machine learning techniques on mobile devices, particularly in the domain of computer vision, in this work we quantitatively study the performance characteristics of 190 real-world vision transformers (ViTs) on mobile devices. Through a comparison with 102 real-world convolutional neural networks (CNNs), we provide insights into the factors that influence the latency of ViT architectures on mobile devices. Based on these insights, we develop a dataset including measured latencies of 1000 synthetic ViTs with representative building blocks and state-of-the-art architectures from two machine learning frameworks and six mobile platforms. Using this dataset, we show that inference latency of new ViTs can be predicted with sufficient accuracy for real-world applications.

Objective: Our aim is to determine if data collected with inertial measurement units (IMUs) during steady-state running could be used to estimate ground reaction forces (GRFs) and to derive biomechanical variables (e.g., contact time, impulse, change in velocity) using lightweight machine-learning approaches. In contrast, state-of-the-art estimation using LSTMs suffers from prohibitive inference times on edge devices, requires expensive training and hyperparameter optimization, and results in black box models. Methods: We proposed a novel lightweight solution, SVD Embedding Regression (SER), using linear regression between SVD embeddings of IMU data and GRF data. We also compared lightweight solutions including SER and k-Nearest-Neighbors (KNN) regression with state-of-the-art LSTMs. Results: We performed extensive experiments to evaluate these techniques under multiple scenarios and combinations of IMU signals and quantified estimation errors for predicting GRFs and biomechanical variables. We did this using training data from different athletes, from the same athlete, or both, and we explored the use of acceleration and angular velocity data from sensors at different locations (sacrum and shanks). Conclusion: Our results illustrated that lightweight solutions such as SER and KNN can be similarly accurate or more accurate than LSTMs. The use of personal data reduced estimation errors of all methods, particularly for most biomechanical variables (as compared to GRFs); moreover, this gain was more pronounced in the lightweight methods. Significance: The study of GRFs is used to characterize the mechanical loading experienced by individuals in movements such as running, which is clinically applicable to identify athletes at risk for stress-related injuries.

Due to the proliferation of inference tasks on mobile devices, state-of-the-art neural architectures are typically designed using Neural Architecture Search (NAS) to achieve good tradeoffs between machine learning accuracy and inference latency. While measuring inference latency of a huge set of candidate architectures during NAS is not feasible, latency prediction for mobile devices is challenging, because of hardware heterogeneity, optimizations applied by machine learning frameworks, and diversity of neural architectures. Motivated by these challenges, we first quantitatively assess the characteristics of neural architectures (specifically, convolutional neural networks for image classification), ML frameworks, and mobile devices that have significant effects on inference latency. Based on this assessment, we propose an operation-wise framework which addresses these challenges by developing operation-wise latency predictors and achieves high accuracy in end-to-end latency predictions, as shown by our comprehensive evaluations on multiple mobile devices using multicore CPUs and GPUs. To illustrate that our approach does not require expensive data collection, we also show that accurate predictions can be achieved on real-world neural architectures using only small amounts of profiling data.

In this paper, we study scheduling of a queueing system with zero knowledge of instantaneous network conditions. We consider a one-hop single-server queueing system consisting of $K$ queues, each with time-varying and non-stationary arrival and service rates. Our scheduling approach builds on an innovative combination of adversarial bandit learning and Lyapunov drift minimization, without knowledge of the instantaneous network state (the arrival and service rates) of each queue. We then present two novel algorithms SoftMW (SoftMaxWeight) and SSMW (Sliding-window SoftMaxWeight), both capable of stabilizing systems that can be stabilized by some (possibly unknown) sequence of randomized policies whose time-variation satisfies a mild condition. We further generalize our results to the setting where arrivals and departures only have bounded moments instead of being deterministically bounded and propose SoftMW+ and SSMW+ that are capable of stabilizing the system. As a building block of our new algorithms, we also extend the classical EXP3.S algorithm for multi-armed bandits to handle unboundedly large feedback signals, which can be of independent interest.

Inference latency prediction on mobile devices is essential for multiple applications, including collaborative inference and neural architecture search. Training accurate latency predictors using ML techniques requires sufficient and representative data; however, collection of such data is challenging. To overcome these challenges, in this work, we focus on constructing a comprehensive dataset that can be used to predict inference latency on mobile devices. Our dataset contains 102 real-world CNNs, 69 real-world ViTs and 1000 synthetic CNNs across 174 diverse experimental environments on mobile platforms, accounting for critical factors affecting inference latency, including hardware heterogeneity, data representations and ML frameworks. Our code is available at: https://github.com/qed-usc/mobile-ml-benchmark.git.

In this paper, we study scheduling of a queueing system with zero knowledge of instantaneous network conditions. We consider a one-hop single-server queueing system consisting of $K$ queues, each with time-varying and non-stationary arrival and service rates. Our scheduling approach builds on an innovative combination of adversarial bandit learning and Lyapunov drift minimization, without knowledge of the instantaneous network state (the arrival and service rates) of each queue. We then present two novel algorithms \texttt{SoftMW} (SoftMaxWeight) and \texttt{SSMW} (Sliding-window SoftMaxWeight), both capable of stabilizing systems that can be stablized by some (possibly unknown) sequence of randomized policies whose time-variation satisfies a mild condition. We further generalize our results to the setting where arrivals and departures only have bounded moments instead of being deterministically bounded and propose \texttt{SoftMW+} and \texttt{SSMW+} that are capable of stabilizing the system. As a building block of our new algorithms, we also extend the classical \texttt{EXP3.S} (Auer et al., 2002) algorithm for multi-armed bandits to handle unboundedly large feedback signals, which can be of independent interest.

Due to the proliferation of inference tasks on mobile devices, state-of-the-art neural architectures are typically designed using Neural Architecture Search (NAS) to achieve good tradeoffs between machine learning accuracy and inference latency. While measuring inference latency of a huge set of candidate architectures during NAS is not feasible, latency prediction for mobile devices is challenging, because of hardware heterogeneity, optimizations applied by machine learning frameworks, and diversity of neural architectures. Motivated by these challenges, we first quantitatively assess the characteristics of neural architectures and mobile devices that have significant effects on inference latency. Based on this assessment, we propose an operation-wise framework which addresses these challenges by developing operation-wise latency predictors and achieves high accuracy in end-to-end latency predictions, as shown by our comprehensive evaluations on multiple mobile devices using multicore CPUs and GPUs. To illustrate that our approach does not require expensive data collection, we also show that accurate predictions can be achieved on real-world neural architectures using only small amounts of profiling data.

Training jobs of deep neural networks (DNNs) can be accelerated through distributed variants of stochastic gradient descent (SGD), where multiple nodes process training examples and exchange updates. The total throughput of the nodes depends not only on their computing power, but also on their networking speeds and coordination mechanism (synchronous or asynchronous, centralized or decentralized), since communication bottlenecks and stragglers can result in sublinear scaling when additional nodes are provisioned. In this paper, we propose two classes of performance models to predict throughput of distributed SGD: fine-grained models , representing many elementary computation/communication operations and their dependencies; and coarse-grained models , where SGD steps at each node are represented as a sequence of high-level phases without parallelism between computation and communication. Using a PyTorch implementation, real-world DNN models and different cloud environments, our experimental evaluation illustrates that, while fine-grained models are more accurate and can be easily adapted to new variants of distributed SGD, coarse-grained models can provide similarly accurate predictions when augmented with ad hoc heuristics, and their parameters can be estimated with profiling information that is easier to collect.

Deep neural networks training jobs and other iterative computations frequently include checkpoints where jobs can be canceled based on the current value of monitored metrics. While most of existing results focus on the performance of all jobs (both successfully completed and canceled), in this work we explore scheduling policies that improve the sojourn time of successful jobs, which are typically more valuable to the user. Our model assumes that each job has a known discrete size distribution (e.g., estimated from previous execution logs) where the largest size value indicates a successful completion, while other size values correspond to termination checkpoints. In the single-server case where all jobs are available for scheduling simultaneously, we prove that optimal schedules do not preempt jobs, even when preemption overhead is negligible. Based on this, we develop a scheduling policy that minimizes the sojourn time of successful jobs asymptotically, i.e., when the number of jobs grows to infinity. Through an extensive numerical study, we show that this policy performs better than existing alternatives even when the number of jobs is finite. For more realistic scenarios with multiple servers and dynamic jobs arrivals, we propose an online approach based on our single-server scheduling policy. Through an extensive simulation study, using real-world traces, we demonstrate that this online approach results in better average sojourn time for successful jobs as compared to existing techniques.

Due to a growing interest in deep learning applications [5], compute-intensive and long-running (hours to days) training jobs have become a significant component of datacenter workloads. A large fraction of these jobs is often exploratory, with the goal of determining the best model structure (e.g., the number of layers and channels in a convolutional neural network), hyperparameters (e.g., the learning rate), and data augmentation strategies for the target application. Notably, training jobs are often terminated early if their learning metrics (e.g., training and validation accuracy) are not converging, with only a few completing successfully. For this motivating application, we consider the problem of scheduling a set of jobs that can be terminated at predetermined checkpoints with known probabilities estimated from historical data. We prove that, in order to minimize the time to complete the first K successful jobs on a single server, optimal scheduling does not require preemption (even when preemption overhead is negligible) and provide an optimal policy; advantages of this policy are quantified through simulation. Related Work. While job scheduling has been investigated extensively in many scenarios (see [6] and [2] for a survey of recent result), most policies require that the cost of waiting times of each job be known at scheduling time; in contrast, in our setting the scheduler does not know which job will be the K-th successful job, and sojourn times of subsequent jobs do not contribute to the target metric. For example, [4, 3] minimize makespan (i.e., the time to complete all jobs) for known execution times and waiting time costs; similarly, Gittins index [1] and SR rank [7] minimize expected sojourn time of all jobs, i.e., both successfully completed jobs and jobs terminated early. Unfortunately, scheduling policies not distinguishing between these two types of jobs may favor jobs where the next stage is short and leads to early termination with high probability, which is an undesirable outcome in our applications of interest.

An accountable algorithmic transparency report (ATR) should ideally investigate (a) transparency of the underlying algorithm, and (b) fairness of the algorithmic decisions, and at the same time preserve data subjects' privacy. However, a provably formal study of the impact to data subjects' privacy caused by the utility of releasing an ATR (that investigates transparency and fairness), has yet to be addressed in the literature. The far-fetched benefit of such a study lies in the methodical characterization of privacy-utility trade-offs for release of ATRs in public, and their consequential application-specific impact on the dimensions of society, politics, and economics. In this paper, we first investigate and demonstrate potential privacy hazards brought on by the deployment of transparency and fairness measures in released ATRs. To preserve data subjects' privacy, we then propose a linear-time optimal-privacy scheme, built upon standard linear fractional programming (LFP) theory, for announcing ATRs, subject to constraints controlling the tolerance of privacy perturbation on the utility of transparency schemes. Subsequently, we quantify the privacy-utility trade-offs induced by our scheme, and analyze the impact of privacy perturbation on fairness measures in ATRs. To the best of our knowledge, this is the first analytical work that simultaneously addresses trade-offs between the triad of privacy, utility, and fairness, applicable to algorithmic transparency reports.

Federated learning allows multiple users to collaboratively train a shared classification model while preserving data privacy. This approach, where model updates are aggregated by a central server, was shown to be vulnerable to poisoning backdoor attacks: a malicious user can alter the shared model to arbitrarily classify specific inputs from a given class. In this article, we analyze the effects of backdoor attacks on federated meta-learning, where users train a model that can be adapted to different sets of output classes using only a few examples. While the ability to adapt could, in principle, make federated learning frameworks more robust to backdoor attacks (when new training examples are benign), we find that even one-shot attacks can be very successful and persist after additional training. To address these vulnerabilities, we propose a defense mechanism inspired by matching networks, where the class of an input is predicted from the similarity of its features with a support set of labeled examples. By removing the decision logic from the model shared with the federation, the success and persistence of backdoor attacks are greatly reduced.

Hybrid cloud architectures, where private clouds or data centers forward part of their workload to public cloud providers to satisfy quality of service (QoS) requirements, are increasingly common due to the availability of on-demand cloud resources that can be provisioned automatically through programming APIs. In this paper, we analyze performance and revenue in federations of hybrid clouds, where private clouds agree to share part of their local computing resources with other members of the federation. Through resource sharing, underprovisioned members can save on public cloud costs, while overprovisioned members can put their idle resources to work. To reward all hybrid clouds for their contributions (computing resources or workload), public cloud savings due to the federation are distributed among members according to Shapley value. We model this cloud architecture with a continuous-time Markov chain and prove that, if all hybrid clouds have the same QoS requirements, their profits are maximized when they join the federation and share all resources. We also show that this result does not hold when hybrid clouds have different QoS requirements, and we provide a solution to evaluate profit for different resource sharing decisions. Finally, our experimental evaluation compares the distribution of public cloud savings according to Shapley value with alternative approaches, illustrating its ability to discourage free riders of the federation.

With the growing workload of inference tasks on mobile devices, state-of-the-art neural architectures (NAs) are typically designed through Neural Architecture Search (NAS) to identify NAs with good tradeoffs between accuracy and efficiency (e.g., latency). Since measuring the latency of a huge set of candidate architectures during NAS is not scalable, approaches are needed for predicting end-to-end inference latency on mobile devices. Such predictions are challenging due to hardware heterogeneity, optimizations applied by ML frameworks, and the diversity of neural architectures. Motivated by these challenges, in this paper, we first quantitatively assess characteristics of neural architectures and mobile devices that have significant effects on inference latency. Based on this assessment, we propose a latency prediction framework which addresses these challenges by developing operation-wise latency predictors, under a variety of settings and a number of hardware devices, with multi-core CPUs and GPUs, achieving high accuracy in end-to-end latency prediction, as shown by our comprehensive evaluations. To illustrate that our approach does not require expensive data collection, we also show that accurate predictions can be achieved on real-world NAs using only small amounts of profiling data.

Artificial neural networks (NNs) in deep learning systems are critical drivers of emerging technologies such as computer vision, text classification, and natural language processing. Fundamental to their success is the development of accurate and efficient NN models. In this article, we report our work on Deep-n-Cheap—an open-source automated machine learning (AutoML) search framework for deep learning models. The search includes both architecture and training hyperparameters and supports convolutional neural networks and multi-layer perceptrons, applicable to multiple domains. Our framework is targeted for deployment on both benchmark and custom datasets, and as a result, offers a greater degree of search space customizability as compared to a more limited search over only pre-existing models from literature. We also introduce the technique of ‘search transfer’, which demonstrates the generalization capabilities of the models found by our framework to multiple datasets. Deep-n-Cheap includes a user-customizable complexity penalty which trades off performance with training time or number of parameters. Specifically, our framework can find models with performance comparable to state-of-the-art while taking 1–2 orders of magnitude less time to train than models from other AutoML and model search frameworks. Additionally, we investigate and develop insight into the search process that should aid future development of deep learning models.

Small ‘boutique’ clouds challenging the big three (AWS, Azure, Google Cloud) on speed, cost, flexibility, on-prem, and hybrid cloud options are slowly on the rise. This paper comments on the work by Pal et al. , in 2020, in relation to the efficiency of practical federated small cloud (SC) resource (e.g., VMs) sharing market structures at a market equilibrium. While the work by Pal et al. , in 2020, guarantees a unique, stable, and efficient sharing equilibrium, it falls short of providing a microscopic view into practically likely sharing network structures (graphs) among SC providers in the market and their effect on market efficiency. Consequently, we envision a graphical federated cloud sharing market and comment on its efficiency guarantee at the market equilibrium. While a symmetric graphical small cloud resource sharing economy is pure-strategy efficient (like in the work by Pal et al. , in 2020) an asymmetric one generates inefficiencies that improves with an increased number of SCs in the sharing market.

We investigate and rationalize how individuals in a surveilled developing economic society value a human-centric data economy (HCDE). We first design and conduct a non-online pilot field experiment on approximately 22500 human subjects across India from 2014–2019, and collect data reflecting the impact of monetary incentives on these subjects to voluntarily trade their (personal) data in the digital surveillance age. Consequently, we study how various degrees of incentive influence subject preferences - both, when they are, or are not well-informed about the commercial malpractices their personal data might be subjected to. We analyze and rationalize two main observations in general for the Indian population: (i) despite being warned of the commercial malpractices associated with their personal data in the mobile and IoT age, they prefer to trade data for incentives, and (ii) the willingness of individuals to trade personal data is statistically heavy-tailed, and hints at following a weak power-law.