Search

Machine learning has emerged as a leading force in revolutionizing technology, education, and almost every aspect of our lives. Reinforcement learning is a sub-field of machine learning that deals with the effects of dynamic feedback and systems that interact with the environment. In these settings, classic statistical and algorithmic guarantees often do not hold because of non i.i.d. data, dynamic feedback, and distribution shift.We develop a framework for single trajectory learning of nonlinear dynamical systems using mixing arguments. Our main result studies the landscape of empirical risk minimization for learning nonlinear dynamical systems from a single trajectory, and provides uniform gradient convergence guarantee, which is combined with novel one-point convexity to facilitate the learning of nonlinear dynamical systems. Our proposed framework allows for non-convex loss landscape and our sample complexity and statistical error rates are optimal in terms of the trajectory length, dimensions of the system and input/noise strength.Next, we study the problem of learning bilinear dynamical systems from a single trajectory of the system’s states and inputs. Our main contribution is the application of martingale small-ball arguments to derive learning guarantees for non-mixing bilinear dynamical systems. We further extend our analysis to time varying dynamical systems by studying the problem of learning non-mixing Markov jump systems. Specifically, we learn the dynamics in each mode and the Markov transition matrix, underlying the evolution of the mode switches, from a single trajectory of the system’s states, inputs, and modes. Our sample complexity and statistical error rates are optimal in terms of the trajectory length, the dimensions of the system and the input/noise strength.Lastly, as a preliminary to the problem of finding the best LTI dynamical system that can minimize least-squares loss given a single trajectory of an unknown dynamical system, we study the simpler problem of finding the best linear model in high dimensions, given a dataset. Specifically, we analyze projected gradient descent algorithm to estimate the population minimizer in the finite sample regime. We show that the nonlinearity of the problem can be treated as uncorrelated noise and establish linear convergence rate and data-dependent estimation error bounds for the projected gradient descent algorithm.

Simulation plays a crucial role in modern academic study, particularly in the field of artificial intelligence (AI). The simulation environment can mimic real-world scenarios, allowing the AI agent to learn, adapt, and make decisions in a controlled and safe setting. This thesis tackles two important problems in building the next generation of artificial general intelligence (AGI): how to efficiently train an AI agent with values and how to overcome the simulation to reality gap to bring the training results to real-world applications. The current studies of AI mainly consider learning about the potential or energy function (U), referring to understanding the impact of the outside environment. The U function helps the agent apprehend the physical world laws, natural potentials, and social norms. However, taking into account the value learning, usually representing modeling one's inside thinking, benefits the agent to derive its goals, intents, and social values. Our research shows that both U and V learning are equally important to the pathway to AGI. The learning of U is usually data-driven. It enables the agent to imitate and complete the task through statistical learning. By incorporating the value function, the agent can spontaneously specify a task plan and its behavior is more in line with human cognition and value.This thesis consists of three parts: (1) Potential function learning, which explores the process of acquiring knowledge or skills that are useful and practical for a particular purpose. (2) Value learning when learning the potential (U) function can not satisfy all the learning goals, which investigates situations where utility-based learning approaches might be limited or ineffective. (3) Combining U and V learning, which focuses on the integration of simulation-based learning and data-driven learning methods. We primarily focus on assessing the effectiveness of U learning within a simulated environment. Our investigation commences with agents operating in a controlled simulated setting, where the action space is intentionally kept small. Through rigorous testing and iterative refinement, we gradually expand the scope of our analysis to encompass agents dealing with increasingly complex and continuous action spaces. Upon achieving compelling results in the simulated realm, we proceed to the crucial next step: transferring the knowledge and expertise gained from the well-trained agents in the simulation space to real-world scenarios. This process entails adapting the learned policies, strategies, and decision-making capabilities of the agents to navigate the intricacies and uncertainties of genuine environments.

Ground Delay Programs (GDP) comprise the main interventions to optimize flight operations in congested air traffic networks. The core GDP objective is to minimize flight delays, but this may result in optimal outcomes for passengers — especially with connecting itineraries. This paper proposes an original passenger-centric approach to GDP by leveraging data on passenger itineraries in flight networks. First, we identify analytical drivers of passenger-centric operations in transportation systems. Second, we develop an integer program that balances flight delays and passenger delays in large-scale GDP operations. A rolling procedure decomposes the problem while ensuring global feasibility, enabling the model’s implementation in short computational times. Third, we propose statistical learning models to predict passenger itineraries and optimize GDP operations accordingly, enabling the model’s implementation when passenger itineraries are unknown by air traffic managers. Computational results based on real-world data suggest that our modeling and computational framework can reduce passenger delays significantly at small increases in flight delay costs, and that these benefits are robust to imperfect knowledge of passenger itineraries. Results highlight two major levers of passenger-centric operations: (i) delay allocation (which flights to delay vs. prioritize), and (ii) delay introduction (whether to deliberately hold flights to avoid passenger misconnections).

In many statistical learning problems, it is desired that the optimal solution conform to an a priori known sparsity structure represented by a directed acyclic graph. Inducing such structures by means of convex regularizers requires nonsmooth penalty functions that exploit group overlapping. Our study focuses on evaluating the proximal operator of the latent overlapping group lasso developed by Jacob et al. in 2009 . We implemented an alternating direction method of multiplier with a sharing scheme to solve large-scale instances of the underlying optimization problem efficiently. In the absence of strong convexity, global linear convergence of the algorithm is established using the error bound theory. More specifically, the paper contributes to establishing primal and dual error bounds when the nonsmooth component in the objective function does not have a polyhedral epigraph. We also investigate the effect of the graph structure on the speed of convergence of the algorithm. Detailed numerical simulation studies over different graph structures supporting the proposed algorithm and two applications in learning are provided. Summary of Contribution: The paper proposes a computationally efficient optimization algorithm to evaluate the proximal operator of a nonsmooth hierarchical sparsity-inducing regularizer and establishes its convergence properties. The computationally intensive subproblem of the proposed algorithm can be fully parallelized, which allows solving large-scale instances of the underlying problem. Comprehensive numerical simulation studies benchmarking the proposed algorithm against five other methods on the speed of convergence to optimality are provided. Furthermore, performance of the algorithm is demonstrated on two statistical learning applications related to topic modeling and breast cancer classification. The code along with the simulation studies and benchmarks are available on the corresponding author’s GitHub website for evaluation and future use.

Background The quality of an emergency department (ED) is highly dependent on its ability to supply efficient, as well as high-quality treatment for all patients. Key performance indicators are important when measuring the performance of an emergency department. This study aimed to perform an exploratory data analysis and to develop an admission prediction model based on a dataset that was constructed from key performance indicators selected by a panel of expert physicians, nurses and hospital administrators. Methods A dataset of 172,695 records was retrospectively collected from an Emergency Department. The relationships within the dataset were analyzed and three machine learning algorithms were compared for an admission predictive model based on the initial patient information. Results The dataset showed that mean length of stay was similar in the different weekdays, there was a positive linear relationship between the length of stay and patient age and the admission predictive model yielded an AUC of 0.79. Conclusions The selected indicators can be used to study whether emergency department allocates its resources properly to cope with overcrowding and the predictive model may be employed by Hospital and ED administrates to fill information gaps and support decision making for the improvement of the key performance indicators.

Increasing evidence demonstrates that observers can learn the likely location of salient singleton distractors during visual search. To date, the reduced attentional capture at high-probability distractor locations has typically been examined using so called compound search, in which by design a target is always present. Here, we explored whether statistical distractor learning can also be observed in a visual detection task, in which participants respond target present if the singleton target is present and respond target absent when the singleton target is absent. If so, this allows us to examine suppression of the location that is likely to contain a distractor both in the presence, but critically also in the absence, of a priority signal generated by the target singleton. In an online variant of the additional singleton paradigm, observers had to indicate whether a unique shape was present or absent, while ignoring a colored singleton, which appeared with a higher probability in one specific location. We show that attentional capture was reduced, but not absent, at high-probability distractor locations, irrespective of whether the display contained a target or not. By contrast, target processing at the high-probability distractor location was selectively impaired on distractor-present displays. Moreover, all suppressive effects were characterized by a gradient such that suppression scaled with the distance to the high-probability distractor location. We conclude that statistical distractor learning can be examined in visual detection tasks, and discuss the implications for attentional suppression due to statistical learning.

Majority voting is often employed as a tool to increase the robustness of data-driven decisions and control policies, a fact which calls for rigorous, quantitative evaluations of the limits and the potentials of majority voting schemes. This letter focuses on the case where the voting agents are binary classifiers and introduces novel bounds on the probability of misclassification conditioned on the size of the majority. We show that these bounds can be much smaller than the traditional upper bounds on the probability of misclassification. These bounds can be used in a 'Probably Approximately Correct' (PAC) setting, which allows for a practical implementation.

Sparse estimation through regularization is gaining popularity in psychological research. Such techniques penalize the complexity of the model and could perform variable/path selection in an automatic way, and thus are particularly useful in models that have small parameter-to-sample-size ratios. This paper gives a detailed tutorial of the R package regsem, which implements regularization for structural equation models. Example R code is also provided to highlight the key arguments of implementing regularized structural equation models in this package. The tutorial ends by discussing remedies of some known drawbacks of a popular type of regularization, computational methods supported by the package that can improve the selection result, and some other practical issues such as dealing with missing data and categorical variables.

Rapid growth in data, computational methods, and computing power is driving a remarkable revolution in what variously is termed machine learning (ML), statistical learning, computational learning, and artificial intelligence. In addition to highly visible successes in machine-based natural language translation, playing the game Go, and self-driving cars, these new technologies also have profound implications for computational and experimental science and engineering, as well as for the exascale computing systems that the Department of Energy (DOE) is developing to support those disciplines. Not only do these learning technologies open up exciting opportunities for scientific discovery on exascale systems, they also appear poised to have important implications for the design and use of exascale computers themselves, including high-performance computing (HPC) for ML and ML for HPC. The overarching goal of the ExaLearn co-design project is to provide exascale ML software for use by Exascale Computing Project (ECP) applications, other ECP co-design centers, and DOE experimental facilities and leadership class computing facilities.

The starting point for the presentation of the similarities and differences between the principles of conducting statistical research according to the rules of both statistical inference and statistical learning is the paradigm theory, formulated by Thomas Kuhn. In the first section of this paper, the essential features of the statistical inference paradigm are characterised, with particular attention devoted to its limitations in contemporary statistical research. Subsequently, the article presents the challenges faced by this research jointly with the expanding opportunities for their effective reduction. The essence of learning from data is discussed and the principles of statistical learning are defined. Moreover, significant features of the statistical learning paradigm are formulated in the context of the differences between the statistical inference paradigm and the statistical learning paradigm. It is emphasised that the statistical learning paradigm, as the more universal one of the two discussed, broadens the possibilities of conducting statistical research, especially in socio-economic sciences.

Previous studies have shown that distributed practice-a training strategy that is known to facilitate memory-is likely to result in greater learning than massed practice. This effect has been demonstrated largely in explicit tasks. The purpose of this study was to test whether statistical learning of artificial grammar is affected by the lag between learning sessions overall, and by high and low complexity stimuli (as measure by chunk strength). Two groups (spaced-short and spaced-long) learned strings of letters created according to a set of rules and were required to produce new strings using given letter sets. For the spaced-short group, the two learning sessions, each including training and a test phase, took place sequentially with a 10-min break, whereas for the spaced-long group, learning sessions were distributed across two days (1-day lag). Overall results showed improved performance following spaced-long practice compared to spaced-short practice. The results also indicated that in the low chunk strength strings (indicating high complexity), both groups demonstrated similar improvement from first to second testing, while in the high chunk strength strings (indicating low complexity), improvement in letter reordering performance was significantly higher when the learning sessions were distributed across two days. This pattern of findings suggests that stimuli complexity affects the extent to which distributed practice enhance artificial grammar learning.

Mobility data, based on global positioning system (GPS) tracking, have been widely used in many areas, such as analyzing travel patterns, investigating transport safety and efficiency, and evaluating travel impacts. Transport modes are essential factors in understanding mobility within the transport system. Therefore, in this study, a significant number of algorithms were tested for transport mode detection. However, no conclusive recommendations can be drawn regarding which method should be used. The evaluation of the performance of the algorithms was not discussed systematically either in current literature. This paper aims to provide an in-depth review of the methods applied in transport mode detection based on GPS tracking data. The performances of the reviewed methods are then compared and evaluated to provide guidance in choosing algorithms for transport mode detection based on GPS tracking data. The results indicate that the majority of current studies are based on a supervised learning method for transport mode detection. Many of the reviewed methods first require manual dataset labeling, which can produce major drawbacks, such as inefficiency and human errors. It was also found that deep learning approaches have the potential to deal with large amounts of unlabeled raw GPS datasets and increase the accuracy and efficiency of transport mode detection.

Several emerging applications call for a fusion of statistical learning and stochastic programming (SP). We introduce a new class of models which we refer to as Predictive Stochastic Programming (PSP). Unlike ordinary SP, PSP models work with datasets which represent random covariates, often refered to as predictors (or features) and responses (or labels) in the machine learning literature. As a result, these PSP models call for methodologies which borrow relevant concepts from both learning and optimization. We refer to such a methodology as Learning Enabled Optimization (LEO). This paper sets forth the foundation for such a framework by introducing several novel concepts such as statistical optimality, hypothesis tests for model-fidelity, generalization error of PSP, and finally, a non-parametric methodology for model selection. These new concepts, which are collectively referred to as LEO, provide a formal framework for modeling, solving, validating, and reporting solutions for PSP models. We illustrate the LEO framework by applying it to a production-marketing coordination model based on combining a pedagogical production planning model with an advertising dataset intended for sales prediction.

It is challenging to model the performance of distributed graph computation. Explicit formulation cannot easily capture the diversified factors and complex interactions in the system. Statistical learning methods require a large number of training samples to generate an accurate prediction model. However, it is time-consuming to run the required graph computation tests to obtain the training samples. In this paper, we propose TransGPerf, a transfer learning based solution that can exploit prior knowledge from a source scenario and utilize a manageable amount of training data for modeling the performance of a target graph computation scenario. Experimental results show that our proposed method is capable of generating accurate models for a wide range of graph computation tasks on PowerGraph and GraphX. It outperforms transfer learning methods proposed for other applications in the literature.

Summarization: Classical geostatistical methods face serious computational challenges if they are confronted with large spatial datasets. The stochastic local interaction (SLI) approach does not require matrix inversion for parameter estimation, spatial prediction, and uncertainty estimation. This leads to better scaling of computational complexity and storage requirements with data size than standard (i.e., without size-reducing modifications) kriging. This contribution presents a simplified SLI model that can handle large data. The SLI method constructs a spatial interaction matrix (precision matrix) that adjusts with minimal user input to the data values, their locations, and sampling density variations. The precision matrix involves compact kernel functions which permit the use of sparse matrix methods. It is proved that the precision matrix of the proposed SLI model is strictly positive definite. In addition, parameter estimation based on likelihood maximization is formulated, and computationally relevant properties of the likelihood function are studied. The interpolation performance of the SLI method is investigated and compared with ordinary kriging using (i) synthetic non-Gaussian data and (ii) coal thickness measurements from approximately 11,500 drill holes (Campbell County, Wyoming, USA). Presented on: Mathematical Geosciences

The coherent organisation of thematic material into large-scale structures within a composition is an important concept in both traditional and cognitive theories of music. However, empirical evide...

Ultra-reliable communication (URC) is a key enabler for supporting immersive and mission-critical 5G applications. Meeting the strict reliability requirements of these applications is challenging due to the absence of accurate statistical models tailored to URC systems. In this letter, the wireless connectivity over dynamic channels is characterized via statistical learning methods. In particular, model-based and data-driven learning approaches are proposed to estimate the non-blocking connectivity statistics over a set of training samples with no knowledge on the dynamic channel statistics. Using principles of survival analysis, the reliability of wireless connectivity is measured in terms of the probability of channel blocking events. Moreover, the maximum transmission duration for a given reliable non-blocking connectivity is predicted in conjunction with the confidence of the inferred transmission duration. Results show that the accuracy of detecting channel blocking events is higher using the model-based method for low to moderate reliability targets requiring low sample complexity. In contrast, the data-driven method shows higher detection accuracy for higher reliability targets at the cost of 100$\times$ sample complexity., Comment: 5 pages, 5 figures, Accepted for IEEE Communications Letters

A recent wave of research has attempted to define fairness quantitatively. In particular, this work has explored what fairness might mean in the context of decisions based on the predictions of statistical and machine learning models. The rapid growth of this new field has led to wildly inconsistent motivations, terminology, and notation, presenting a serious challenge for cataloging and comparing definitions. This article attempts to bring much-needed order. First, we explicate the various choices and assumptions made—often implicitly—to justify the use of prediction-based decision-making. Next, we show how such choices and assumptions can raise fairness concerns and we present a notationally consistent catalog of fairness definitions from the literature. In doing so, we offer a concise reference for thinking through the choices, assumptions, and fairness considerations of prediction-based decision-making.

Application Traffic Identification is an imperative device for sorting out the system as it is the most popular approach to distinguish and characterize the network traffic created from different applications. The classification using conventional Port-based and Payload-based techniques has become counterproductive due to inconsistencies. However, in recent times, approaches with machine learning and statistical techniques have guaranteed higher accuracy. However, learning techniques are inadequate for solving problems with Time and Memory intricacies in vast datasets. Hence, the proposed paper presents a novel scheme of Statistical based traffic classification named Multi-Phased Statistical Based Classification methodology that renders Semi-supervised machines with advanced K-medoid clustering and C5.0 Classification algorithm. The proposed system displays a classic competence in observing the known and unknown application flows by statistical features utilization scheme that enhances the classification preciseness. Further, the trial results show that the proposed work outperforms previous approaches by achieving a higher granularity of 98–99% and reducing complexities. Ultimately, the new proposed work is evaluated on our campus traffic traces (AU-IDS). It is proven that the proposed approach accomplishes a higher exactness rate and thus encourages its implementation in real-time.

A training image free, high-order sequential simulation method is proposed herein, which is based on the efficient inference of high-order spatial statistics from the available sample data. A statistical learning framework in kernel space is adopted to develop the proposed simulation method. Specifically, a new concept of aggregated kernel statistics is proposed to enable sparse data learning. The conditioning data in the proposed high-order sequential simulation method appear as data events corresponding to the attribute values associated with the so-called spatial templates of various geometric configurations. The replicates of the data events act as the training data in the learning framework for inference of the conditional probability distribution and generation of simulated values. These replicates are mapped into spatial Legendre moment kernel spaces, and the kernel statistics are computed thereafter, encapsulating the high-order spatial statistics from the available data. To utilize the incomplete information from the replicates, which partially match the spatial template of a given data event, the aggregated kernel statistics combine the ensemble of the elements in different kernel subspaces for statistical inference, embedding the high-order spatial statistics of the replicates associated with various spatial templates into the same kernel subspace. The aggregated kernel statistics are incorporated into a learning algorithm to obtain the target probability distribution in the underlying random field, while preserving in the simulations the high-order spatial statistics from the available data. The proposed method is tested using a synthetic dataset, showing the reproduction of the high-order spatial statistics of the sample data. The comparison with the corresponding high-order simulation method using TIs emphasizes the generalization capacity of the proposed method for sparse data learning.

Considering the escalating rates of exhaustion of non-renewable energy resources, coupled with the harmful environmental side effects of harnessing them (e.g. damage to public health via air pollut...

The brain is highly sensitive to auditory regularities and exploits the predictable order of sounds in many situations, from parsing complex auditory scenes, to the acquisition of language. To understand the impact of stimulus predictability on perception, it is important to determine how the detection of predictable structure influences processing and attention. Here, we use pupillometry to gain insight into the effect of sensory regularity on arousal. Pupillometry is a commonly used measure of salience and processing effort, with more perceptually salient or perceptually demanding stimuli consistently associated with larger pupil diameters. In two experiments we tracked human listeners' pupil dynamics while they listened to sequences of 50-ms tone pips of different frequencies. The order of the tone pips was either random, contained deterministic (fully predictable) regularities (experiment 1, n = 18, 11 female) or had a probabilistic regularity structure (experiment 2, n = 20, 17 female). The sequences were rapid, preventing conscious tracking of sequence structure thus allowing us to focus on the automatic extraction of different types of regularities. We hypothesized that if regularity facilitates processing by reducing processing demands, a smaller pupil diameter would be seen in response to regular relative to random patterns. Conversely, if regularity is associated with heightened arousal and attention (i.e., engages processing resources) the opposite pattern would be expected. In both experiments we observed a smaller sustained (tonic) pupil diameter for regular compared with random sequences, consistent with the former hypothesis and confirming that predictability facilitates sequence processing. SIGNIFICANCE STATEMENT The brain is highly sensitive to auditory regularities. To appreciate the impact that the presence of predictability has on perception, we need to better understand how a predictable structure influences processing and attention. We recorded listeners' pupil responses to sequences of tones that followed either a predictable or unpredictable pattern, as the pupil can be used to implicitly tap into these different cognitive processes. We found that the pupil showed a smaller sustained diameter to predictable sequences, indicating that predictability eased processing rather than boosted attention. The findings suggest that the pupil response can be used to study the automatic extraction of regularities, and that the effects are most consistent with predictability helping the listener to efficiently process upcoming sounds.

To provide a comprehensive analysis of small molecule genotoxic potential we have developed and validated an automated, high-content, high throughput, image-based in vitro Micronucleus (IVM) assay. This assay simultaneously assesses micronuclei and multiple additional cellular markers associated with genotoxicity. Acoustic dosing (≤ 2 mg) of compound is followed by a 24-h treatment and a 24-h recovery period. Confocal images are captured [Cell Voyager CV7000 (Yokogawa, Japan)] and analysed using Columbus software (PerkinElmer). As standard the assay detects micronuclei (MN), cytotoxicity and cell-cycle profiles from Hoechst phenotypes. Mode of action information is primarily determined by kinetochore labelling in MN (aneugencity) and γH2AX foci analysis (a marker of DNA damage). Applying computational approaches and implementing machine learning models alongside Bayesian classifiers allows the identification of, with 95% accuracy, the aneugenic, clastogenic and negative compounds within the data set (Matthews correlation coefficient: 0.9), reducing analysis time by 80% whilst concurrently minimising human bias. Combining high throughput screening, multiparametric image analysis and machine learning approaches has provided the opportunity to revolutionise early Genetic Toxicology assessment within AstraZeneca. By multiplexing assay endpoints and minimising data generation and analysis time this assay enables complex genotoxicity safety assessments to be made sooner aiding the development of safer drug candidates.

© 2020, The Psychonomic Society, Inc.People are sensitive to regularities in the environment. Recent studies employing the additional singleton paradigm showed that a singleton distractor that appeared more often in one specific location than in all other locations may lead to attentional suppression of high-probability distractor locations. This in turn effectively reduced the attentional capture effect by the salient distractor singleton. However, in basically all of these previous studies, the probability that the salient distractor was presented at this specific location was relatively high (i.e., 65%; or a ratio of 13:1 between high- and low-probability locations). The question we addressed here was whether participants still can learn the regularities in the display even when these regularities are quite subtle. We systematically manipulated the ratio of the distractor appearing at the high- and low-probability location from 2:1 to 8:1. We asked the question whether the suppression effect would depend on the probabilities of the distractor appearing in the high-probability location. The results showed that the suppression of the high-probability location was linearly related to the high-low-probability ratio. In other words, the more evidence that a distractor appears more often at a particular location, the stronger the suppression. This indicates that the distribution of attention is optimally adapted to the statistical regularities present in the display.

This study explores online insurance ventures by customers’ behavioral profiles from data mining and visualization. First, we employ a simple decision-tree statistical learning method on unique hand-collected and processed data in Taiwan. This method considers first-time online insurance subscribers as new ones without a proper marketing connection after six months. Second, using advanced clustering technique and decision tree statistical learning on first-time purchasing customers, we find they repurchase online travel insurance for different purposes varying periods. Finally, we get robust results engaging different segmentations of customer data. In these ways, we enable marketing strategies to work with generating decision rules.

Melt pool temperature contains abundant information on metallurgical and mechanical aspects of products produced by additive manufacturing. Forecasting melt pool temperature profile during a process can help in reducing microstructural porosity and residual stresses. Although analytical and numerical models were reported, the performance of these are questionable when applied in real-time. Hence, we developed data-driven models to address this challenge, for continuous forecasting layer-wise melt pool temperature using a hybrid deep learning technique. The melt pool temperature forecasting by the proposed CNN-LSTM model is found to be better than other benchmark models in terms of accuracy and efficiency. The model results have shown that combining CNN and LSTM networks can extract the spatial and temporal information from the melt pool temperature data. Further, the proposed model results are compared with existing statistical and machine learning models. The performance measures of the proposed CNN-LSTM model indicate a greater potential for in-situ monitoring of additive manufacturing process.