Downloads & Free Reading Options - Results

In this dissertation, a series of studies are designed to investigate the brain connectivity patterns (both functional connectivity and effective connectivity), comparing subjects with depression history vs. control and subjects with OCD vs. control. Depression and OCD are examples of trans-diagnostic disorders, which may be more likely due to alteration of connectivity patterns (ex., hyper-connectivity in default mode network or other networks) rather than localized brain abnormality.

It is well known in the literature that the problem of learning the structure of Bayesian networks is very hard to tackle: its computational complexity is super-exponential in the number of nodes in the worst case and polynomial in most real-world scenarios. Efficient implementations of score-based structure learning benefit from past and current research in optimisation theory, which can be adapted to the task by using the network score as the objective function to maximise. This is not true for approaches based on conditional independence tests, called constraint-based learning algorithms. The only optimisation in widespread use, backtracking, leverages the symmetries implied by the definitions of neighbourhood and Markov blanket. In this paper we illustrate how backtracking is implemented in recent versions of the bnlearn R package, and how it degrades the stability of Bayesian network structure learning for little gain in terms of speed. As an alternative, we describe a software architecture and framework that can be used to parallelise constraint-based structure learning algorithms (also implemented in bnlearn) and we demonstrate its performance using four reference networks and two real-world data sets from genetics and systems biology. We show that on modern multi-core or multiprocessor hardware parallel implementations are preferable over backtracking, which was developed when single-processor machines were the norm.

The problem of learning the structure of a high dimensional graphical model from data has received considerable attention in recent years. In many applications such as sensor networks and proteomics it is often expensive to obtain samples from all the variables involved simultaneously. For instance, this might involve the synchronization of a large number of sensors or the tagging of a large number of proteins. To address this important issue, we initiate the study of a novel graphical model selection problem, where the goal is to optimize the total number of scalar samples obtained by allowing the collection of samples from only subsets of the variables. We propose a general paradigm for graphical model selection where feedback is used to guide the sampling to high degree vertices, while obtaining only few samples from the ones with the low degrees. We instantiate this framework with two specific active learning algorithms, one of which makes mild assumptions but is computationally expensive, while the other is more computationally efficient but requires stronger (nevertheless standard) assumptions. Whereas the sample complexity of passive algorithms is typically a function of the maximum degree of the graph, we show that the sample complexity of our algorithms is provable smaller and that it depends on a novel local complexity measure that is akin to the average degree of the graph. We finally demonstrate the efficacy of our framework via simulations.

Threats from the internet, particularly malicious software (i.e., malware) often use cryptographic algorithms to disguise their actions and even to take control of a victim's system (as in the case of ransomware). Malware and other threats proliferate too quickly for the time-consuming traditional methods of binary analysis to be effective. By automating detection and classification of cryptographic algorithms, we can speed program analysis and more efficiently combat malware. This thesis will present several methods of leveraging machine learning to automatically discover and classify cryptographic algorithms in compiled binary programs. While further work is necessary to fully evaluate these methods on real-world binary programs, the results in this paper suggest that machine learning can be used successfully to detect and identify cryptographic primitives in compiled code. Currently, these techniques successfully detect and classify cryptographic algorithms in small single-purpose programs, and further work is proposed to apply them to real-world examples.

In this study we gathered extracted information from different databases and studied the performance of machine learning algorithms in detecting skull base tumors via CT & MRI

In this study we gathered extracted information from different databases and studied the performance of machine learning algorithms in detecting skull base tumors via CT & MRI

This interim report describes work completed from December 1988 to November 1991. The original purpose of the research program funded by this grant was to study self-organized systems which adapt and learn. The originally proposed research fell into two main categories: (1). Biological Models of Self- Organization; and (2). New Self-Organized Learning Systems. During the period for which this grant was funded, significant progress was made in both areas. In addition, some new areas related to neural network learning are being explored as an outgrowth of the original proposal. These include: (1). Model Selection Techniques; and (2). Estimating Generalization Performance.

This interim report describes work completed from December 1988 to November 1991. The original purpose of the research program funded by this grant was to study self-organized systems which adapt and learn. The originally proposed research fell into two main categories: (1). Biological Models of Self- Organization; and (2). New Self-Organized Learning Systems. During the period for which this grant was funded, significant progress was made in both areas. In addition, some new areas related to neural network learning are being explored as an outgrowth of the original proposal. These include: (1). Model Selection Techniques; and (2). Estimating Generalization Performance.

This paper presents new and effective algorithms for learning kernels. In particular, as shown by our empirical results, these algorithms consistently outperform the so-called uniform combination solution that has proven to be difficult to improve upon in the past, as well as other algorithms for learning kernels based on convex combinations of base kernels in both classification and regression. Our algorithms are based on the notion of centered alignment which is used as a similarity measure between kernels or kernel matrices. We present a number of novel algorithmic, theoretical, and empirical results for learning kernels based on our notion of centered alignment. In particular, we describe efficient algorithms for learning a maximum alignment kernel by showing that the problem can be reduced to a simple QP and discuss a one-stage algorithm for learning both a kernel and a hypothesis based on that kernel using an alignment-based regularization. Our theoretical results include a novel concentration bound for centered alignment between kernel matrices, the proof of the existence of effective predictors for kernels with high alignment, both for classification and for regression, and the proof of stability-based generalization bounds for a broad family of algorithms for learning kernels based on centered alignment. We also report the results of experiments with our centered alignment-based algorithms in both classification and regression.

We consider the problem of learning the structure of Ising models (pairwise binary Markov random fields) from i.i.d. samples. While several methods have been proposed to accomplish this task, their relative merits and limitations remain somewhat obscure. By analyzing a number of concrete examples, we show that low-complexity algorithms often fail when the Markov random field develops long-range correlations. More precisely, this phenomenon appears to be related to the Ising model phase transition (although it does not coincide with it).

Smoothed analysis is a framework for analyzing the complexity of an algorithm, acting as a bridge between average and worst-case behaviour. For example, Quicksort and the Simplex algorithm are widely used in practical applications, despite their heavy worst-case complexity. Smoothed complexity aims to better characterize such algorithms. Existing theoretical bounds for the smoothed complexity of sorting algorithms are still quite weak. Furthermore, empirically computing the smoothed complexity via its original definition is computationally infeasible, even for modest input sizes. In this paper, we focus on accurately predicting the smoothed complexity of sorting algorithms, using machine learning techniques. We propose two regression models that take into account various properties of sorting algorithms and some of the known theoretical results in smoothed analysis to improve prediction quality. We show experimental results for predicting the smoothed complexity of Quicksort, Mergesort, and optimized Bubblesort for large input sizes, therefore filling the gap between known theoretical and empirical results.

One of the classical problems in machine learning and data mining is feature selection. A feature selection algorithm is expected to be quick, and at the same time it should show high performance. MeLiF algorithm effectively solves this problem using ensembles of ranking filters. This article describes two different ways to improve MeLiF algorithm performance with parallelization. Experiments show that proposed schemes significantly improves algorithm performance and increase feature selection quality.

Gravitational wave astronomy has become a reality after the historical detections accomplished during the first observing run of the two advanced LIGO detectors. In the following years, the number of detections is expected to increase significantly with the full commissioning of the advanced LIGO, advanced Virgo and KAGRA detectors. The development of sophisticated data analysis techniques to improve the opportunities of detection for low signal-to-noise-ratio events is hence a most crucial effort. We present in this paper one such technique, dictionary-learning algorithms, which have been extensively developed in the last few years and successfully applied mostly in the context of image processing. However, to the best of our knowledge, such algorithms have not yet been employed to denoise gravitational wave signals. By building dictionaries from numerical relativity templates of both, binary black holes mergers and bursts of rotational core collapse, we show how machine-learning algorithms based on dictionaries can be also successfully applied for gravitational wave denoising. We use a subset of signals from both catalogs, embedded in non-white Gaussian noise, to assess our techniques with a large sample of tests and to find the best model parameters. The application of our method to the actual signal GW150914 shows promising results. Dictionary-learning algorithms could be a complementary addition to the gravitational wave data analysis toolkit. They may be used to extract signals from noise and to infer physical parameters if the data are in good enough agreement with the morphology of the dictionary atoms.

A Stroke is a disease when there is insufficient blood supply to the brain, which causes cell death. It is currently the world’s biggest cause of death. Upon examining the affected individuals, a number of risk variables that are thought to be connected to the cause of stroke have been identified. Numerous studies have been conducted to predict and categorize stroke disorders using the risk variables. The majority of the models are built using machine learning and data mining technologies. In this work, we have employed four machine learning algorithms to identify the type of stroke that may have happened based on medical report data and an individual’s physicalcondition. We have gathered a sizable amount of hospital entries. This study employs many methodologies, including decision trees, Naive Bayes, ANN algorithm, and Random Forest algorithm. Thus, the aim of this study is to evaluate the mentioned algorithmsand determine which one does the task more accurately. After completing all of the evaluations, we can conclude that the Random Forest method has the highest accuracy of all the algorithms with 99%.

"Classical" First Order (FO) algorithms of convex optimization, such as Mirror Descent algorithm or Nesterov's optimal algorithm of smooth convex optimization, are well known to have optimal (theoretical) complexity estimates which do not depend on the problem dimension. However, to attain the optimality, the domain of the problem should admit a "good proximal setup". The latter essentially means that 1) the problem domain should satisfy certain geometric conditions of "favorable geometry", and 2) the practical use of these methods is conditioned by our ability to compute at a moderate cost {\em proximal transformation} at each iteration. More often than not these two conditions are satisfied in optimization problems arising in computational learning, what explains why proximal type FO methods recently became methods of choice when solving various learning problems. Yet, they meet their limits in several important problems such as multi-task learning with large number of tasks, where the problem domain does not exhibit favorable geometry, and learning and matrix completion problems with nuclear norm constraint, when the numerical cost of computing proximal transformation becomes prohibitive in large-scale problems. We propose a novel approach to solving nonsmooth optimization problems arising in learning applications where Fenchel-type representation of the objective function is available. The approach is based on applying FO algorithms to the dual problem and using the {\em accuracy certificates} supplied by the method to recover the primal solution. While suboptimal in terms of accuracy guaranties, the proposed approach does not rely upon "good proximal setup" for the primal problem but requires the problem domain to admit a Linear Optimization oracle -- the ability to efficiently maximize a linear form on the domain of the primal problem.

This is a follow-up sensitivity study on r-mode gravitational wave signals from newborn neutron stars illustrating the applicability of machine learning algorithms for the detection of long-lived gravitational-wave transients. In this sensitivity study we examine three machine learning algorithms (MLAs): artificial neural networks (ANNs), support vector machines (SVMs) and constrained subspace classifiers (CSCs). The objective of this study is to compare the detection efficiency that MLAs can achieve with the efficiency of conventional detection algorithms discussed in an earlier paper. Comparisons are made using 2 distinct r-mode waveforms. For the training of the MLAs we assumed that some information about the distance to the source is given so that the training was performed over distance ranges not wider than half an order of magnitude. The results of this study suggest that machine learning algorithms are suitable for the detection of long-lived gravitational-wave transients and that when assuming knowledge of the distance to the source, MLAs are at least as efficient as conventional methods.

We provide the first oracle efficient sublinear regret algorithms for adversarial versions of the contextual bandit problem. In this problem, the learner repeatedly makes an action on the basis of a context and receives reward for the chosen action, with the goal of achieving reward competitive with a large class of policies. We analyze two settings: i) in the transductive setting the learner knows the set of contexts a priori, ii) in the small separator setting, there exists a small set of contexts such that any two policies behave differently in one of the contexts in the set. Our algorithms fall into the follow the perturbed leader family \cite{Kalai2005} and achieve regret $O(T^{3/4}\sqrt{K\log(N)})$ in the transductive setting and $O(T^{2/3} d^{3/4} K\sqrt{\log(N)})$ in the separator setting, where $K$ is the number of actions, $N$ is the number of baseline policies, and $d$ is the size of the separator. We actually solve the more general adversarial contextual semi-bandit linear optimization problem, whilst in the full information setting we address the even more general contextual combinatorial optimization. We provide several extensions and implications of our algorithms, such as switching regret and efficient learning with predictable sequences.

We study algorithms for online linear optimization in Hilbert spaces, focusing on the case where the player is unconstrained. We develop a novel characterization of a large class of minimax algorithms, recovering, and even improving, several previous results as immediate corollaries. Moreover, using our tools, we develop an algorithm that provides a regret bound of $\mathcal{O}\Big(U \sqrt{T \log(U \sqrt{T} \log^2 T +1)}\Big)$, where $U$ is the $L_2$ norm of an arbitrary comparator and both $T$ and $U$ are unknown to the player. This bound is optimal up to $\sqrt{\log \log T}$ terms. When $T$ is known, we derive an algorithm with an optimal regret bound (up to constant factors). For both the known and unknown $T$ case, a Normal approximation to the conditional value of the game proves to be the key analysis tool.

The problem of distributed learning and channel access is considered in a cognitive network with multiple secondary users. The availability statistics of the channels are initially unknown to the secondary users and are estimated using sensing decisions. There is no explicit information exchange or prior agreement among the secondary users. We propose policies for distributed learning and access which achieve order-optimal cognitive system throughput (number of successful secondary transmissions) under self play, i.e., when implemented at all the secondary users. Equivalently, our policies minimize the regret in distributed learning and access. We first consider the scenario when the number of secondary users is known to the policy, and prove that the total regret is logarithmic in the number of transmission slots. Our distributed learning and access policy achieves order-optimal regret by comparing to an asymptotic lower bound for regret under any uniformly-good learning and access policy. We then consider the case when the number of secondary users is fixed but unknown, and is estimated through feedback. We propose a policy in this scenario whose asymptotic sum regret which grows slightly faster than logarithmic in the number of transmission slots.

A systematic review and meta-analysis of the diagnostic test accuracy of deep learning in diagnosing or grading pterygium.

A systematic review and meta-analysis of the diagnostic test accuracy of deep learning in diagnosing or grading pterygium.

Rumination is associated with the onset, duration and severity of a depression. Being distracted from ruminative thoughts (“distraction hypothesis”) is discussed as a possible mechanism of action of the antidepressant effect of moderate to vigorous exercise, which is well-established (Heissel et al., 2023; Morres et al., 2019). In this project, we decode rumination from electroencephalography (EEG) data using machine learning algorithms. Decoded rumination and self-reports are used to predict possible changes in rumination through exercise. Decoded rumination provides a more objective measure of rumination, additional to and beyond self-reports, that might be less biased and shed light into the underlying neurophysiological correlates of rumination. In this project, we will investigate whether moderate-intensity exercise (ME) reduces rumination compared to a sedentary control condition (SED). ME will be performed as continuous exercise at 100-110% of the individual first lactate threshold. In the sedentary control condition, participants sit inactive in a chair. Each condition is performed for 30 minutes. Participants will complete a single factor (ME vs. SED) within-subject design in randomised order while EEG is measured. EEG is applied with 59 electrodes according to the 10-20 system. Additionally, data is measured from 4 EOG electrodes, 1 electrode at muscle risorius, 4 bipolar electrodes at muscle trapezius and 4 bipolar electrodes at sternocleidomastoid muscle. In a previous part of the project (https://doi.org/10.17605/OSF.IO/C5JF9), decoders (i.e., support-vector classification models) are trained to predict rumination (versus distraction) from EEG data during experimentally induced rumination or distraction. In the current project, the trained decoders predict the class (i.e., rumination vs. distraction) and class probability of rumination from continuous EEG data features (i.e., alpha and theta power across the 59 channels and a connectivity matrix between all channels) measured during the exercises. The class probability for rumination is analysed in 7.5 s data segments across the time course of ME or SED, respectively. Furthermore, self-reported rumination will be assessed before and after each condition using the Perseverative Thinking Questionnaire state (PTQ-S; Ehring et al., 2011) and during the conditions using visual analogue scales (VAS). We hypothesize that the mean change of self-reported rumination as well as the mean decoded probability of rumination is significantly lower in the ME condition compared to the SED condition. By implementing a novel, more objective measurement of rumination in combination with validated and well-established self-reports, this project will help to understand whether distraction mediates the antidepressant effect of exercise.

Genetic learning algorithms are widely used to control ultrafast optical pulse shapes for photo-induced quantum control of atoms and molecules. An unresolved issue is how to use the solutions found by these algorithms to learn about the system's quantum dynamics. We propose a simple method based on covariance analysis of the control space, which can reveal the degrees of freedom in the effective control Hamiltonian. We have applied this technique to stimulated Raman scattering in liquid methanol. A simple model of two-mode stimulated Raman scattering is consistent with the results.

SOL is an open-source library for scalable online learning algorithms, and is particularly suitable for learning with high-dimensional data. The library provides a family of regular and sparse online learning algorithms for large-scale binary and multi-class classification tasks with high efficiency, scalability, portability, and extensibility. SOL was implemented in C++, and provided with a collection of easy-to-use command-line tools, python wrappers and library calls for users and developers, as well as comprehensive documents for both beginners and advanced users. SOL is not only a practical machine learning toolbox, but also a comprehensive experimental platform for online learning research. Experiments demonstrate that SOL is highly efficient and scalable for large-scale machine learning with high-dimensional data.

We consider distributed optimization over orthogonal collision channels in spatial random access networks. Users are spatially distributed and each user is in the interference range of a few other users. Each user is allowed to transmit over a subset of the shared channels with a certain attempt probability. We study both the non-cooperative and cooperative settings. In the former, the goal of each user is to maximize its own rate irrespective of the utilities of other users. In the latter, the goal is to achieve proportionally fair rates among users. Simple distributed learning algorithms are developed to solve these problems. The efficiencies of the proposed algorithms are demonstrated via both theoretical analysis and simulation results.

Although stochastic approximation learning methods have been widely used in the machine learning literature for over 50 years, formal theoretical analyses of specific machine learning algorithms are less common because stochastic approximation theorems typically possess assumptions which are difficult to communicate and verify. This paper presents a new stochastic approximation theorem for state-dependent noise with easily verifiable assumptions applicable to the analysis and design of important deep learning algorithms including: adaptive learning, contrastive divergence learning, stochastic descent expectation maximization, and active learning.

Perma.cc archive of http://tradingtuitions.com/machine-learning-algorithms-trading created on 2020-09-27 20:46:04+00:00.

The objective of this paper is to find an alternative to conventional method of concrete mix design. For finding the alternative, 4 machine learning algorithms viz. multi-variable linear regression, Support Vector Regression, Decision Tree Regression and Artificial Neural Network for designing concrete mix of desired properties. The multi-variable linear regression model is just a simplistic baseline model, support vector regression Artificial Neural Network model were made because past researchers worked heavily on them, Decision tree model was made by authors own intuition. Their results have been compared to find the best algorithm. Finally, we check if the best performing algorithm is accurate enough to replace the convention method. For this, we utilize the concrete mix designs done in lab for various on site designs. The models have been designed for both mixes types – with plasticizer and without plasticizer The paper presents detailed comparison of four models Based on the results obtained from the four models, the best one has been selected based on high accuracy and least computational cost. Each sample had 24 features initially, out of which, most significant features were chosen which were contributing towards prediction of a variable using f regression and p values and models were trained on those selected features. Based on the R squared value, best fitting models were selected among the four algorithms used. From the paper, the author(s) conclude that decision tree regression is best for calculating the amount of ingredients required with R squared values close to 0.8 for most of the models. DTR model is also computationally cheaper than ANN and future works with DTR in mix design is highly recommended in this paper.

The present study aims to perform a systematic literature mapping to identify studies that address the use of machine learning algorithms for predicting various outcomes in children and adolescents diagnosed with COVID-19.

Shortly following arrest, judicial officers must decide whether to detain the arrested person in jail or to release him or her back into the community while awaiting trial. This is an extremely important decision in a criminal case (Bechtel, Holsinger, Lowenkamp, Warren, 2017). This decision relates to later case decisions (e.g., Ulmer, 2012; Kutateladze, Andiloro, Johnson, & Spohn, 2014), case outcomes (Oleson, Lowenkamp, Wooldredge, VanNostrand, & Cadigan, 2015), as well as outcomes even after a case is disposed (Cadigan & Lowenkamp, 2011; Lowenkamp, VanNostrand, & Holsinger, 2013). For example, those detained pre-trial are much more likely than those released pre-trial to plead guilty (Patterson & Lynch, 1991; Sutton, 2013), to be convicted of a felony (Schlesinger, 2007), and to receive a longer final sentence (Sacks & Ackerman, 2012). According to McCoy (2007), the decision to detain or release someone pre-trial is so critical that it determines mostly everything in a criminal case. Both the American Bar Association (2002) and the National Association of Pre-trial Services Agencies (2004) strongly recommend the use of an objective and research-based pre-trial risk assessment instrument to assist judicial officers’ in making this decision. One goal of these instruments is to identify people who are likely to recidivate. Researchers and practitioners have developed various pre-trial risk assessment instruments within the last two decades. Some prominent examples include the Virginia Pre-trial Risk Assessment Instrument (VPRAI) developed by the Virginia Department of Criminal Justice Services (VanNostrand, 2003), and the Public Safety Assessment (PSA) developed by the Laura and John Arnold Foundation (Lowenkamp, VanNostrand, & Holsinger, 2013; VanNostrand & Lowenkamp, 2013). Desmarais, Zottola, Clarke, and Lowder (2020) reviewed several risk assessment instruments including the VPRAI and PSA and found that they predicted recidivism with good to excellent accuracy. For example, the VPRAI discriminated those who had new arrests during the pre-trial period from those who did not 64 – 69% of the time. Moreover, these instruments were similarly predictive across racial and ethnic groups and were similarly predictive for both men and women. Still, Desmarais et al. emphasized the need for continued investigation of the predictive accuracy of pre-trial risk assessment instruments. Once judicial officers decide to release someone back into the community, they either release that person under supervision or without any supervision. Pre-trial supervision comes with certain conditions and restrictions that can include periodic check-ins with a case manager, maintaining employment, ankle monitoring, alcohol testing and treatment, and cognitive behavioral therapy (Clarke, 1988; Mamalian, 2011; VanNostrand & Keebler, 2009; VanNostrand, Rose, & Weibrecht, 2011). Those released pre-trial may also have access to social services that include opportunities to take part in education programs or employment training as well as transitional housing. The goal of pre-trial supervision is to provide an alternative to detention while minimizing recidivism and failures to appear in court. It is not clear whether pre-trial supervision reduces recidivism more so than simply releasing people pre-trial. The available research on the effectiveness of pre-trial supervision is limited. Bechtel et al. (2017) conducted a meta-analysis of 16 studies that investigated the impact of various pre-trial supervision conditions (e.g., ankle monitoring) on recidivism and found that none of the conditions reduced recidivism. Bechtel et al. tempered these findings by emphasizing that the quality of the research included in the meta analysis was poor and that the field of pre-trial research lacks methodological rigor. For example, they state, “the quality of the research that could be included in the current analysis was not very good” (p. 460). Elsewhere, they note that, “it is striking that although more than 800 potential studies on pre-trial were identified, less than 20% contained data, and the percentage of studies with the information necessary to synthesize the findings into a meta-analytic review was even lower than 20%” (p. 459). In fact, of the 16 studies that were included in the meta-analysis only four studies were peer-reviewed. They also added that there is a “great need for new and more rigorous pre-trial research in all related areas” (p. 459). They conclude by calling for researchers to “conduct methodologically rigorous studies that are submitted to peer-reviewed journals” (p. 463). Here, we answer the calls from Desmarais et al. (2020) and Bechtel et al. (2017). First, we compare the predictive accuracy of the VPRAI and the PSA. Second, we use two modern machine-learning-based matching algorithms to determine the causal impact of pre-trial supervision on recidivism. These algorithms, called Fast Large-scale Almost Matching Exactly -- FLAME (Wang et al. (2021) and Dynamic Almost Matching Exactly -- DAME (Liu, Dieng, Roy, Rudin, and Volfovsky (2019), stem from a new causal inference framework called Almost Matching Exactly, or Learning-to-Match. We discuss these machine-learning algorithms in detail below, but in general, they match people who receive pre-trial supervision to similar individuals who were released pre-trial. The virtue of this machine-learning matching process is that it establishes causality and can therefore test whether pre-trial supervision has a causal effect on recidivism. Next, we detail our confirmatory and exploratory hypotheses.

Autism spectrum disorder (ASD), also known as autism, is a complex neurodevelopmental disorder that occurs early in development and affects children's development to varying degrees. It could lead to persistent deficits in children's social communication skills and behavior (K et al.,2017). Early prediction of the occurrence of the disease can provide assistance for standardized diagnosis and early intervention by pediatric healthcare physicians, pediatric general practitioners, and developmental behavioral pediatricians. It is estimated that about 1 in every 100 children worldwide has autism (Zeidan et al.,2022). The prevalence of ASD in children aged 2-16 in China is about 0.70%, and there are over 2 million ASD children under 14 years old. Some scholars believe that the actual number may reach 2.6-8 million people (Jieqiong Zhang & Chen,2021). The large ASD population has become one of the major public health issues (Lord et al.,2018). The guidelines developed by the American Psychiatric Association, the Diagnostic and Statistical Manual of Mental Disorders (DSM-5), point out that ASD patients have deficits in restricted interests and repetitive behaviors, which limits their communication and social interaction abilities (Publishing,2013). Due to the lack of awareness of this disease among parents, they may overlook the early signs exhibited by their children. Therefore, it is essential to make early predictions about it (Anurekha G & P,2017). Machine learning is a constantly evolving field of research aimed at building precise predictive models from their respective research datasets. It includes search methods, artificial intelligence, mathematical modeling, and other predictive elements. Machine learning is an automated tool that does not require much human involvement in data processing, and this computational learning algorithm is increasingly applied in research on neurocognitive disorders (Bone et al.,2015). This article reviews the research on machine learning algorithms for early prediction of ASD, analyzes the features and model characteristics included in the literature, in order to provide reference for early diagnosis of ASD.

In the construction field, compressive strength is one of the most critical parameters of concrete. However, a significant amount of physical effort and natural raw materials are required to produce concrete. In addition, the curing period of concrete for at least 28 days is a must for attaining the required compressive strength. Various types of industrial and agricultural wastes have been used in concrete to reduce cement consumption and problems due to its production. Therefore, considering such constraints, the application of Artificial Intelligence (AI) has been widely used in the current scenarios to predict the desired output parameters. In the present study, 12 input parameters have been considered along with 455 data points and nine Machine Learning (ML) models to forecast the compressive strength of Fly Ash (FA) based concrete. The output from the models has been compared to find the best-fit model in terms of numerous analyses such as visual descriptive statistics, errors,  R 2 , Taylor’s diagram, Feature Importance (FI), and scatter plots. Based on the analysis of the current study, Decision Tree (DT) and Gradient Boost (GB) were found to be the best-fit model because of the least errors and higher  R 2  values as compared to other models.

We provide new results concerning noise-tolerant and sample-efficient learning algorithms under $s$-concave distributions over $\mathbb{R}^n$ for $-\frac{1}{2n+3}\le s\le 0$. The new class of $s$-concave distributions is a broad and natural generalization of log-concavity, and includes many important additional distributions, e.g., the Pareto distribution and $t$-distribution. This class has been studied in the context of efficient sampling, integration, and optimization, but much remains unknown concerning the geometry of this class of distributions and their applications in the context of learning. The challenge is that unlike the commonly used distributions in learning (uniform or more generally log-concave distributions), this broader class is not closed under the marginalization operator and many such distributions are fat-tailed. In this work, we introduce new convex geometry tools to study the properties of s-concave distributions and use these properties to provide bounds on quantities of interest to learning including the probability of disagreement between two halfspaces, disagreement outside a band, and disagreement coefficient. We use these results to significantly generalize prior results for margin-based active learning, disagreement-based active learning, and passively learning of intersections of halfspaces. Our analysis of geometric properties of s-concave distributions might be of independent interest to optimization more broadly.

We provide new results concerning noise-tolerant and sample-efficient learning algorithms under $s$-concave distributions over $\mathbb{R}^n$ for $-\frac{1}{2n+3}\le s\le 0$. The new class of $s$-concave distributions is a broad and natural generalization of log-concavity, and includes many important additional distributions, e.g., the Pareto distribution and $t$-distribution. This class has been studied in the context of efficient sampling, integration, and optimization, but much remains unknown concerning the geometry of this class of distributions and their applications in the context of learning. The challenge is that unlike the commonly used distributions in learning (uniform or more generally log-concave distributions), this broader class is not closed under the marginalization operator and many such distributions are fat-tailed. In this work, we introduce new convex geometry tools to study the properties of s-concave distributions and use these properties to provide bounds on quantities of interest to learning including the probability of disagreement between two halfspaces, disagreement outside a band, and disagreement coefficient. We use these results to significantly generalize prior results for margin-based active learning, disagreement-based active learning, and passively learning of intersections of halfspaces. Our analysis of geometric properties of s-concave distributions might be of independent interest to optimization more broadly.

We present an approach for learning simple algorithms such as copying, multi-digit addition and single digit multiplication directly from examples. Our framework consists of a set of interfaces, accessed by a controller. Typical interfaces are 1-D tapes or 2-D grids that hold the input and output data. For the controller, we explore a range of neural network-based models which vary in their ability to abstract the underlying algorithm from training instances and generalize to test examples with many thousands of digits. The controller is trained using $Q$-learning with several enhancements and we show that the bottleneck is in the capabilities of the controller rather than in the search incurred by $Q$-learning.

As the military continues to expand its use of intelligent agents in a variety of operational aspects, event prediction and learning algorithms are becoming more and more important. In this paper, we conduct a detailed analysis of two such algorithms: Variable Order Markov and Look-Up Table models. Each model employs different parameters for prediction, and this study attempts to determine which model is more accurate in its prediction and why. We find the models contrast in that the Variable Order Markov Model increases its average prediction probability, our primary performance measure, with increased maximum model order, while the Look-Up Table Model decreases average prediction probability with increased recency time threshold. In addition, statistical tests of results of each model indicate a consistency in each model's prediction capabilities, and most of the variation in the results could be explained by model parameters.

Kevin Ellis is an assistant professor at Cornell and currently a research scientist at Common Sense Machines. His research focuses on artificial intelligence, program synthesis, and neurosymbolic models.Kevin's PhD thesis is titled "Algorithms for Learning to Induce Programs", which he completed in 2020 at MIT. We discuss Kevin's work at the intersection of machine learning and program induction, including inferring graphics programs from images and drawings, DreamCoder, and more.Episode notes: https://cs.nyu.edu/~welleck/episode26.htmlFollow the Thesis Review (@thesisreview) and Sean Welleck (@wellecks) on Twitter, and find out more info about the show at https://cs.nyu.edu/~welleck/podcast.htmlSupport The Thesis Review at www.patreon.com/thesisreview or www.buymeacoffee.com/thesisreview

Talk given by Tom Dean, of Google. Given to the Redwood Center for Theoretical Neuroscience at UC Berkeley on January 20, 2010.. Abstract. Graphics processors (GPUs) and massively-multi-core architectures are becoming more powerful, less costly and more energy efficient, and the related programming language issues are beginning to sort themselves out. That said most researchers don’t want to be writing code that depends on any particular architecture or parallel programming model. Linear algebra, Fourier analysis and image processing have standard libraries that are being ported to exploit SIMD parallelism in GPUs. We can depend on the massively-multiple-core machines du jour to support these libraries and on the high-performance-computing (HPC) community to do the porting for us or with us. These libraries can significantly accelerate important applications in image processing, data analysis and information retrieval. We can develop APIs and the necessary run-time support so that code relying on these libraries will run on any machine in a cluster of computers but exploit GPUs whenever available. This strategy allows us to move toward hybrid computing models that enable a wider range of opportunities for parallelism without requiring the special training of programmers or the disadvantages of developing code that depends on specialized hardware or programming models. This talk summarizes the state of the art in massively-multi-core architectures, presents experimental results that demonstrate the potential for significant performance gains in the two general areas of image processing and machine learning, provides examples of the proposed programming interface, and some more detailed experimental results on one particular problem involving video-content analysis.

Talk given by Tom Dean, of Google. Given to the Redwood Center for Theoretical Neuroscience at UC Berkeley on January 20, 2010.. Abstract. Graphics processors (GPUs) and massively-multi-core architectures are becoming more powerful, less costly and more energy efficient, and the related programming language issues are beginning to sort themselves out. That said most researchers don’t want to be writing code that depends on any particular architecture or parallel programming model. Linear algebra, Fourier analysis and image processing have standard libraries that are being ported to exploit SIMD parallelism in GPUs. We can depend on the massively-multiple-core machines du jour to support these libraries and on the high-performance-computing (HPC) community to do the porting for us or with us. These libraries can significantly accelerate important applications in image processing, data analysis and information retrieval. We can develop APIs and the necessary run-time support so that code relying on these libraries will run on any machine in a cluster of computers but exploit GPUs whenever available. This strategy allows us to move toward hybrid computing models that enable a wider range of opportunities for parallelism without requiring the special training of programmers or the disadvantages of developing code that depends on specialized hardware or programming models. This talk summarizes the state of the art in massively-multi-core architectures, presents experimental results that demonstrate the potential for significant performance gains in the two general areas of image processing and machine learning, provides examples of the proposed programming interface, and some more detailed experimental results on one particular problem involving video-content analysis.

We study two mixed robust/average-case submodular partitioning problems that we collectively call Submodular Partitioning. These problems generalize both purely robust instances of the problem (namely max-min submodular fair allocation (SFA) and min-max submodular load balancing (SLB) and also generalize average-case instances (that is the submodular welfare problem (SWP) and submodular multiway partition (SMP). While the robust versions have been studied in the theory community, existing work has focused on tight approximation guarantees, and the resultant algorithms are not, in general, scalable to very large real-world applications. This is in contrast to the average case, where most of the algorithms are scalable. In the present paper, we bridge this gap, by proposing several new algorithms (including those based on greedy, majorization-minimization, minorization-maximization, and relaxation algorithms) that not only scale to large sizes but that also achieve theoretical approximation guarantees close to the state-of-the-art, and in some cases achieve new tight bounds. We also provide new scalable algorithms that apply to additive combinations of the robust and average-case extreme objectives. We show that these problems have many applications in machine learning (ML). This includes: 1) data partitioning and load balancing for distributed machine algorithms on parallel machines; 2) data clustering; and 3) multi-label image segmentation with (only) Boolean submodular functions via pixel partitioning. We empirically demonstrate the efficacy of our algorithms on real-world problems involving data partitioning for distributed optimization of standard machine learning objectives (including both convex and deep neural network objectives), and also on purely unsupervised (i.e., no supervised or semi-supervised learning, and no interactive segmentation) image segmentation.

In dictionary learning, also known as sparse coding, the algorithm is given samples of the form $y = Ax$ where $x\in \mathbb{R}^m$ is an unknown random sparse vector and $A$ is an unknown dictionary matrix in $\mathbb{R}^{n\times m}$ (usually $m > n$, which is the overcomplete case). The goal is to learn $A$ and $x$. This problem has been studied in neuroscience, machine learning, visions, and image processing. In practice it is solved by heuristic algorithms and provable algorithms seemed hard to find. Recently, provable algorithms were found that work if the unknown feature vector $x$ is $\sqrt{n}$-sparse or even sparser. Spielman et al. \cite{DBLP:journals/jmlr/SpielmanWW12} did this for dictionaries where $m=n$; Arora et al. \cite{AGM} gave an algorithm for overcomplete ($m >n$) and incoherent matrices $A$; and Agarwal et al. \cite{DBLP:journals/corr/AgarwalAN13} handled a similar case but with weaker guarantees. This raised the problem of designing provable algorithms that allow sparsity $\gg \sqrt{n}$ in the hidden vector $x$. The current paper designs algorithms that allow sparsity up to $n/poly(\log n)$. It works for a class of matrices where features are individually recoverable, a new notion identified in this paper that may motivate further work. The algorithm runs in quasipolynomial time because they use limited enumeration.

There exist many algorithms for learning how to play repeated bimatrix games. Most of these algorithms are justified in terms of some sort of theoretical guarantee. On the other hand, little is known about the empirical performance of these algorithms. Most such claims in the literature are based on small experiments, which has hampered understanding as well as the development of new multiagent learning (MAL) algorithms. We have developed a new suite of tools for running multiagent experiments: the MultiAgent Learning Testbed (MALT). These tools are designed to facilitate larger and more comprehensive experiments by removing the need to build one-off experimental code. MALT also provides baseline implementations of many MAL algorithms, hopefully eliminating or reducing differences between algorithm implementations and increasing the reproducibility of results. Using this test suite, we ran an experiment unprecedented in size. We analyzed the results according to a variety of performance metrics including reward, maxmin distance, regret, and several notions of equilibrium convergence. We confirmed several pieces of conventional wisdom, but also discovered some surprising results. For example, we found that single-agent $Q$-learning outperformed many more complicated and more modern MAL algorithms.

In recent years, several authors have used probabilistic graphical models to learn expression modules and their regulatory programs from gene expression data. Here, we demonstrate the use of the synthetic data generator SynTReN for the purpose of testing and comparing module network learning algorithms. We introduce a software package for learning module networks, called LeMoNe, which incorporates a novel strategy for learning regulatory programs. Novelties include the use of a bottom-up Bayesian hierarchical clustering to construct the regulatory programs, and the use of a conditional entropy measure to assign regulators to the regulation program nodes. Using SynTReN data, we test the performance of LeMoNe in a completely controlled situation and assess the effect of the methodological changes we made with respect to an existing software package, namely Genomica. Additionally, we assess the effect of various parameters, such as the size of the data set and the amount of noise, on the inference performance. Overall, application of Genomica and LeMoNe to simulated data sets gave comparable results. However, LeMoNe offers some advantages, one of them being that the learning process is considerably faster for larger data sets. Additionally, we show that the location of the regulators in the LeMoNe regulation programs and their conditional entropy may be used to prioritize regulators for functional validation, and that the combination of the bottom-up clustering strategy with the conditional entropy-based assignment of regulators improves the handling of missing or hidden regulators.

Information retrieval using probabilistic techniques has attracted significant attention on the part of researchers in information and computer science over the past few decades. In the 1980s, knowledge-based techniques also made an impressive contribution to intelligent information retrieval and indexing. More recently, information science researchers have turned to other newer inductive learning techniques including symbolic learning, genetic algorithms, and simulated annealing. These newer techniques, which are grounded in diverse paradigms, have provided great opportunities for researchers to enhance the information processing and retrieval capabilities of current information systems. In this article, we first provide an overview of these newer techniques and their use in information systems. In this article, we first provide an overview of these newer techniques and their use in information retrieval research. In order to familiarize readers with the techniques, we present three promising methods: The symbolic ID3 algorithm, evolution-based genetic algorithms, and simulated annealing. We discuss their knowledge representations and algorithms in the unique context of information retrieval. An experiment using a 8000-record COMPEN database was performed to examine the performances of these inductive query-by-example techniques in comparison with the performance of the conventional relevance feedback method. The machine learning techniques were shown to be able to help identify new documents which are similar to documents initially suggested by users, and documents which contain similar concepts to each other. Genetic algorithms, in particular, were found to out-perform relevance feedback in both document recall and precision.

Two far-reaching trends in computing have grown in significance in recent years. First, statistical machine learning has entered the mainstream as a broadly useful tool set for building applications. Second, the need to protect systems against malicious adversaries continues to increase across computing applications. The growing intersection of these trends compels us to investigate how well machine learning performs under adversarial conditions. When a learning algorithm succeeds in adversarial conditions, it is an algorithm for secure learning. The crucial task is to evaluate the resilience of learning systems and determine whether they satisfy requirements for secure learning. In this thesis, we show that the space of attacks against machine learning has a structure that we can use to build secure learning systems. This thesis makes three high-level contributions. First, we develop a framework for analyzing attacks against machine learning systems. We present a taxonomy that describes the space of attacks against learning systems, and we model such attacks as a cost-sensitive game between the attacker and the defender. We survey attacks in the literature and describe them in terms of our taxonomy. Second, we develop two concrete attacks against a popular machine learning spam filter and present experimental results confirming their effectiveness. These attacks demonstrate that real systems using machine learning are vulnerable to compromise. Third, we explore defenses against attacks with both a high-level discussion of defenses within our taxonomy and a multi-level defense against attacks in the domain of virus detection. Using both global and local information, our virus defense successfully captures many viruses designed to evade detection. Our framework, exploration of attacks, and discussion of defenses provides a strong foundation for constructing secure learning systems.

Two far-reaching trends in computing have grown in significance in recent years. First, statistical machine learning has entered the mainstream as a broadly useful tool set for building applications. Second, the need to protect systems against malicious adversaries continues to increase across computing applications. The growing intersection of these trends compels us to investigate how well machine learning performs under adversarial conditions. When a learning algorithm succeeds in adversarial conditions, it is an algorithm for secure learning. The crucial task is to evaluate the resilience of learning systems and determine whether they satisfy requirements for secure learning. In this thesis, we show that the space of attacks against machine learning has a structure that we can use to build secure learning systems. This thesis makes three high-level contributions. First, we develop a framework for analyzing attacks against machine learning systems. We present a taxonomy that describes the space of attacks against learning systems, and we model such attacks as a cost-sensitive game between the attacker and the defender. We survey attacks in the literature and describe them in terms of our taxonomy. Second, we develop two concrete attacks against a popular machine learning spam filter and present experimental results confirming their effectiveness. These attacks demonstrate that real systems using machine learning are vulnerable to compromise. Third, we explore defenses against attacks with both a high-level discussion of defenses within our taxonomy and a multi-level defense against attacks in the domain of virus detection. Using both global and local information, our virus defense successfully captures many viruses designed to evade detection. Our framework, exploration of attacks, and discussion of defenses provides a strong foundation for constructing secure learning systems.

The educational sector faced many types of research in predicting student performance based on supervised and unsupervised machine learning algorithms. Most students' performance data are imbalanced, where the final classes are not equally represented. Besides the size of the dataset, this problem affects the model's prediction accuracy. In this paper, the Synthetic Minority Oversampling TEchnique (SMOTE) filter is applied to the dataset to find its effect on the model's accuracy. Four feature selection approaches are applied to find the most correlated attributes that affect the students' performance. The SMOTE filter is examined before and after applying feature selection approaches to measure the model's accuracy with supervised and unsupervised algorithms. Three supervised/unsupervised algorithms are examined based on feature selection approaches to predict the students' performance. The findings show that supervised algorithms (logistic model trees (LMT), simple logistic, and random forest) got high accuracy after applying SMOTE without feature selection. The prediction accuracies of unsupervised algorithms (Canopy, expectations maximization (EM), and farthest first) are enhanced after applying feature selection approaches and SMOTE filter. 

The educational sector faced many types of research in predicting student performance based on supervised and unsupervised machine learning algorithms. Most students' performance data are imbalanced, where the final classes are not equally represented. Besides the size of the dataset, this problem affects the model's prediction accuracy. In this paper, the Synthetic Minority Oversampling TEchnique (SMOTE) filter is applied to the dataset to find its effect on the model's accuracy. Four feature selection approaches are applied to find the most correlated attributes that affect the students' performance. The SMOTE filter is examined before and after applying feature selection approaches to measure the model's accuracy with supervised and unsupervised algorithms. Three supervised/unsupervised algorithms are examined based on feature selection approaches to predict the students' performance. The findings show that supervised algorithms (logistic model trees (LMT), simple logistic, and random forest) got high accuracy after applying SMOTE without feature selection. The prediction accuracies of unsupervised algorithms (Canopy, expectations maximization (EM), and farthest first) are enhanced after applying feature selection approaches and SMOTE filter. 

This paper shows that if one is provided with a loss function, it can be used in a natural way to specify a distance measure quantifying the similarity of any two supervised learning algorithms, even non-parametric algorithms. Intuitively, this measure gives the fraction of targets and training sets for which the expected performance of the two algorithms differs significantly. Bounds on the value of this distance are calculated for the case of binary outputs and 0-1 loss, indicating that any two learning algorithms are almost exactly identical for such scenarios. As an example, for any two algorithms A and B, even for small input spaces and training sets, for less than 2e(-50) of all targets will the difference between A's and B's generalization performance of exceed 1%. In particular, this is true if B is bagging applied to A, or boosting applied to A. These bounds can be viewed alternatively as telling us, for example, that the simple English phrase 'I expect that algorithm A will generalize from the training set with an accuracy of at least 75% on the rest of the target' conveys 20,000 bytes of information concerning the target. The paper ends by discussing some of the subtleties of extending the distance measure to give a full (non-parametric) differential geometry of the manifold of learning algorithms.

This paper shows that if one is provided with a loss function, it can be used in a natural way to specify a distance measure quantifying the similarity of any two supervised learning algorithms, even non-parametric algorithms. Intuitively, this measure gives the fraction of targets and training sets for which the expected performance of the two algorithms differs significantly. Bounds on the value of this distance are calculated for the case of binary outputs and 0-1 loss, indicating that any two learning algorithms are almost exactly identical for such scenarios. As an example, for any two algorithms A and B, even for small input spaces and training sets, for less than 2e(-50) of all targets will the difference between A's and B's generalization performance of exceed 1%. In particular, this is true if B is bagging applied to A, or boosting applied to A. These bounds can be viewed alternatively as telling us, for example, that the simple English phrase 'I expect that algorithm A will generalize from the training set with an accuracy of at least 75% on the rest of the target' conveys 20,000 bytes of information concerning the target. The paper ends by discussing some of the subtleties of extending the distance measure to give a full (non-parametric) differential geometry of the manifold of learning algorithms.