Bayesian networks (BNs) are a widely used graphical model in machine learning. As learning the structure of BNs is NP-hard, high-performance computing methods are necessary for constructing large-scale networks. In th...
详细信息
Bayesian networks (BNs) are a widely used graphical model in machine learning. As learning the structure of BNs is NP-hard, high-performance computing methods are necessary for constructing large-scale networks. In this article, we present a parallel framework to scale BN structure learning algorithms to tens of thousands of variables. Our framework is applicable to learning algorithms that rely on the discovery of Markov blankets (MBs) as an intermediate step. We demonstrate the applicability of our framework by parallelizing three different algorithms: Grow-Shrink (GS), Incremental Association MB (IAMB), and Interleaved IAMB (Inter-IAMB). Our implementations are available as part of an open-source software called ramBLe, and are able to construct BNs from real data sets with tens of thousands of variables and thousands of observations in less than a minute on 1024 cores, with a speedup of up to 845X and 82.5% efficiency. Furthermore, we demonstrate using simulated data sets that our proposed parallel framework can scale to BNs of even higher dimensionality. Our implementations were selected for the reproducibility challenge component of the 2021 student cluster competition (SCC'21), which tasked undergraduate teams from around the world with reproducing the results that we obtained using the implementations. We discuss details of the challenge and the results of the experiments conducted by the top teams in the competition. The results of these experiments indicate that our key results are reproducible, despite the use of completely different data sets and experiment infrastructure, and validate the scalability of our implementations.
Validation and Verification (V&V) of Artificial Intelligence (AI) based cyber physical systems such as Autonomous Vehicles (AVs) is currently a vexing and unsolved problem. AVs integrate subsystems in areas such a...
详细信息
Validation and Verification (V&V) of Artificial Intelligence (AI) based cyber physical systems such as Autonomous Vehicles (AVs) is currently a vexing and unsolved problem. AVs integrate subsystems in areas such as detection, sensor fusion, localization, perception, and path planning. Each of these subsystems contains significant AI content integrated with traditional hardware and software components. The complexity for validating even a subsystem is daunting and the task of validating the whole system is nearly impossible. Fundamental research in advancing the state-of-the-art for AV V&V is required. However, for V&V researchers, it is exceedingly difficult to make progress because of the massive infrastructure requirements to demonstrate the viability of any solution. This paper presents PolyVerif, the world's first open-source solution focused on V&V researchers with the objective of accelerating the state-of-the-art for AV V&V research. PolyVerif provides an AI design and verification framework consisting of a digital twin creation process, an open-source AV engine, access to several open-source physics based simulators, and open-source symbolic test generation engines. PolyVerif's objective is to arm V&V researchers with a framework which extends the state-of-the-art on any one of the many major axes of interest and use the remainder of the infrastructure to quickly demonstrate the viability of their solution. Given its open-source nature, researchers can also contribute their innovations to the project. Using this critical property of open-source environments, the innovation rate of the whole research community to solve these vexing issues can be greatly accelerated. Finally, the paper also presents results from several projects which have used PolyVerif.
Decentralized stochastic gradient algorithms efficiently solve large-scale finite-sum optimization problems when all agents in the network are reliable. However, most of these algorithms are not resilient to adverse c...
详细信息
Decentralized stochastic gradient algorithms efficiently solve large-scale finite-sum optimization problems when all agents in the network are reliable. However, most of these algorithms are not resilient to adverse conditions, such as malfunctioning agents, software bugs, and cyber attacks. This article aims to handle a class of general composite optimization problems over multiagent systems (MASs) in the presence of an unknown number of Byzantine agents. Building on a resilient aggregation mechanism and the proximal-gradient mapping method, a Byzantine-resilient decentralized stochastic proximal-gradient algorithmic framework is proposed, dubbed Prox-DBRO-VR, which achieves an optimization and control goal using only local computations and communications. To asymptotically reduce the noise variance arising from local gradient estimation and accelerate the convergence, we incorporate two localized variance-reduced (VR) techniques (SAGA and LSVRG) into Prox-DBRO-VR to design Prox-DBRO-SAGA and Prox-DBRO-LSVRG. By analyzing the contraction relationships among the gradient-learning error, resilient consensus condition, and convergence error in a unified theoretical framework, it is proved that both Prox-DBRO-SAGA and Prox-DBRO-LSVRG, with a well-designed constant (resp., decaying) step-size, converge linearly (resp., sublinearly) inside an error ball around the optimal solution to the original problem under standard assumptions. A tradeoff between convergence accuracy and Byzantine resilience in both linear and sublinear cases is also characterized. In numerical experiments, the effectiveness and practicability of the proposed algorithms are manifested via resolving a decentralized sparse machine learning problem under various Byzantine attacks.
The proposed motion cueing algorithm (MCA), based on a reinforcement learning algorithm using gradient information to directly update the control policy, introduces three significant enhancements. First, transform the...
详细信息
The proposed motion cueing algorithm (MCA), based on a reinforcement learning algorithm using gradient information to directly update the control policy, introduces three significant enhancements. First, transform the complex simulator environment into a differentiable simulator environment that provides gradient information at each time step and use this gradient information to directly update the control policy. Second, the network architecture is reconfigured into a concurrent controller format, similar to Model Predictive Control (MPC). This controller processes a sequence of vehicle motion reference signals over a future period, utilizing a multi-layer perceptron to generate the simulator's motion reference control signal sequences for the same duration. Unlike the online optimization employed in MPC, this algorithm as an offline optimization method, providing substantial computational advantages when integrated into the driving simulator. As the prediction horizon increases, the algorithm demonstrates superior computational efficiency, which helps reduce the incidence of motion sickness during the use of the driving simulator. Third, a loss function specifically designed for the motion simulator is proposed. This function incorporates constraints derived from the MPC framework to address workspace limitations and applies them to workspace management. These constraints restrict the platform's acceleration and speed near the workspace boundaries, allowing for better utilization of the available space. The algorithm is validated using Carla's autonomous driving simulation software as the dataset generator. During the training process, the proposed algorithm in this paper achieves an order-of-magnitude improvement in convergence speed compared to conventional training methods of PPO and DDPG. Simulations with a 10-step prediction horizon indicate that the Root Mean Square Error (RMSE) produced by this algorithm is comparable to that of the MCA based on MPC (MPC-MC
Real-time power system algorithms are necessary for grid advancement, but few practical applications have been demonstrated in a research usability context. The work in this paper consists of implementing a data corre...
详细信息
Real-time power system algorithms are necessary for grid advancement, but few practical applications have been demonstrated in a research usability context. The work in this paper consists of implementing a data correction algorithm, its deployment within realistic substation equipment, and the design of a testbed to demonstrate the overall framework and its usability. A digital simulator is used to generate Phasor Measurement Unit (PMU) data for the algorithm. The algorithm corrects data that has been perturbed by GPS spoofing attacks. Finally, the entire system is visualized on power-utility software, SEL Synchrowave Operations. Considerations and potential issues are discussed and are applicable to digital Hardware-in-the-Loop (HIL) systems as well as to field-deployed systems. The system is demonstrated with 11 simultaneously GPS-spoofing attacked PMUs in a 21 PMU system. The HIL testbed developed in this paper provides a valuable tool for easily testing a variety of real-time power system algorithms and the communications and control necessary for them operate successfully.
The shuffle operations are the bottleneck when mapping the FFT-like algorithms to the vector single instruction multiple data(SIMD) *** propose six(three pairs) innovative vector memoryaccess shuffle fused instruction...
详细信息
The shuffle operations are the bottleneck when mapping the FFT-like algorithms to the vector single instruction multiple data(SIMD) *** propose six(three pairs) innovative vector memoryaccess shuffle fused instructions, which have been proved mathematically. Combined with the proposed modified binary-exchange method, the innovative instructions can efficiently address the bottleneck problem for decimationin-frequency or decimation-in-time(DIF/DIT) radix-2/4FFT-like algorithms, reach a performance improvement by 17.9%–111.2% and reduce the code size by 5.4%–39.8%.In addition, the proposed instructions fit some hybridradix FFTs and are suitable for the terms of the initial or result data placement for general algorithms. The software and hardware costs of the proposed instructions are moderate.
Detecting and understanding reasons for defects and inadvertent behavior in software is challenging due to their increasing complexity. In configurable software systems, the combinatorics that arises from the multitud...
详细信息
ISBN:
(纸本)9781450392211
Detecting and understanding reasons for defects and inadvertent behavior in software is challenging due to their increasing complexity. In configurable software systems, the combinatorics that arises from the multitude of features a user might select from adds a further layer of complexity. We introduce the notion of feature causality, which is based on counterfactual reasoning and inspired by the seminal definition of actual causality by Halpern and Pearl. Feature causality operates at the level of system configurations and is capable of identifying features and their interactions that are the reason for emerging functional and non-functional properties. We present various methods to explicate these reasons, in particular well-established notions of responsibility and blame that we extend to the feature-oriented setting. Establishing a close connection of feature causality to prime implicants, we provide algorithms to effectively compute feature causes and causal explications. By means of an evaluation on a wide range of configurable software systems, including community benchmarks and real-world systems, we demonstrate the feasibility of our approach: We illustrate how our notion of causality facilitates to identify root causes, estimate the effects of features, and detect feature interactions.
The selection of the most suitable franchisee applicant in an uncertain environment in a particular moment of time is a key decision for a franchisor and the success of a franchising business. In this work, for the fi...
详细信息
ISBN:
(数字)9788396242396
ISBN:
(纸本)9788396242396
The selection of the most suitable franchisee applicant in an uncertain environment in a particular moment of time is a key decision for a franchisor and the success of a franchising business. In this work, for the first time, we describe a problem for choosing the optimal candidate for the franchise chain and algorithm for a solution in terms of temporal intuitionistic fuzzy pairs and index matrices as a means for data analysis in uncertain conditions over time. We also use our software utility to demonstrate the proposed algorithm and to apply the decision support approach to a franchisee selection for the largest fast food restaurant chain in Bulgaria.
For the decades that followed the publishing of the Cooper-Harper report that formalized a standard and universally recognized handling qualities pilot rating scale, researchers have sought to correlate pilot compensa...
详细信息
ISBN:
(数字)9781624107115
ISBN:
(纸本)9781624107115
For the decades that followed the publishing of the Cooper-Harper report that formalized a standard and universally recognized handling qualities pilot rating scale, researchers have sought to correlate pilot compensation-as well as physical and mental workload-with the assigned rating. A quantitative correlation remains elusive. In recent years, new physiological measurement devices have been developed that together with software processing tools can provide accurate measures of psychophysiological measures including cognitive workload, distraction, and high/low engagement based on electroencephalogram (EEG) and electrocardiogram (ECG) measures (i.e., brain waves and heart rate variability). The pilot compensation referred to in the Cooper-Harper scale is also a function of task performance measures that reflect aircraft characteristics and inceptor activity that reflects upon physical workload. Using a new piloted simulation test database generated in Manned Flight Simulator's containerized rotary-wing simulator with ten experienced test pilots, a machine learning-based software algorithm that integrates a disparate mix of pilot-vehicle system, physiological, and task performance measures was used to further develop an approach to predict handling qualities levels and ratings.
暂无评论