Labeling challenging behaviors demonstrated in Applied Behavioral Analysis therapy is essential in treating challenging behaviors in individuals with Autism Spectrum Disorder. The understanding and accuracy of labeling these challenging behaviors correlate with the efficacy of behavioral therapy. This paper further improves a quantitative solution from previous work by applying neural document embeddings with Doc2Vec to represent the clinical notes on 1,917 recorded instances of challenging behaviors from therapy sessions. Similarly, we implemented the document embeddings as training data into binary and multiclass classification models, whose accuracies ranged between 82.9% and 99.3%. The results with the new input demonstrated that document neural embeddings provided a more accurate representation of the clinical notes, thus providing a more accurate identification of challenging behaviors.

Climate impacts are particularly noticeable for the nations that share the Nile basin with an increase in hotter temperatures and fluctuating precipitation which expands natural catastrophes. Provincial work is required to precisely predict floods and dry seasons, thus preparing, and adapting to climatic events to build climate resilience among these Nile basin nations. In this context, an index indicating vulnerability to drought is derived for the Nile basin using Geographically Weighted Principal Component Analysis (GWPCA) and K-means clustering. Several climate indicators images related to atmosphere, land, and ocean are collected to build clusters categorized as high, mild, and low drought risk. Additionally, STL decomposition is conducted for the Palmer Drought Severity Index (PDSI) using time series data from 2010–2020 for the Nile basin to identify exceptional drought events for the past decade. Furthermore, correlations among PDSI and other climate indicators are analyzed using time series.

The understanding and treatment of challenging behaviors in individuals with Autism Spectrum Disorder are paramount to enabling the success of behavioral therapy; an essential step in this process is the labeling of challenging behaviors demonstrated in therapy sessions. This paper seeks to add quantitative depth to this otherwise qualitative task of challenging behavior classification. Here we leverage neural document embeddings with Word2Vec to represent clinical notes capturing 1,917 recorded instance of challenging behaviors from therapy sessions conducted by a large autism treatment provider. These embeddings then serve as training data for supervised machine learning algorithms in both binary and multiclass classification tasks to identify challenging behaviors, achieving high classification accuracies ranging from 82.7% to 98.5%. We demonstrate that the semantic queues derived from the language of challenging behavior descriptions, modeled using natural language processing techniques, can be successfully leveraged to extract and identify challenging behaviors from real-world clinical data.

Land cover changes in the Mediterranean region have caused phenomenal shifts in weather patterns in areas of Europe and North Africa over the past years. Complicated short-term and gradual land cover changes in this region are caused by natural and human factors. Remote sensing and GIS techniques are widely used to detect the changes in land use land cover (LULC). Additionally, statistical analysis, data mining, and machine learning models are implemented in literature studies. In this study, a Siamese Neural Network, an extension of Convolution Neural Networks (CNN), is used to detect the landcover type changes that occurred in the Mediterranean region from 2009 to 2018 using MCD12Q1.006 data. Similarity scores are generated in Siamese Neural Network using 2009 and 2018 landcover classified images. Based on the similarity scores change of landcover classes are identified. Finally, base models such as Nearest Neighbor model and a Confusion matrix which is generated from image histograms are used to compare the performance of the Siamese Neural Network.

This study uses eye-tracking experiment data to predict the fixation points for children with Autism Spectrum Disorder (ASD) and Typically Developing (TD) for 14 ASD and 14 TD subjects for 300 scenic images [1]. Based on explanatory Logistic Regression models, it is evident that fixation patterns for both ASD and TD subjects focus on the center of each scenic image. Using gradient boosting the researchers successfully identify 31.7% and 39.5% of all fixation points in the top decile of predicted fixation points for ASD and TD subjects respectively. Results conclude that TD subjects have less variability in their eye movement and fixation points leading to increased accuracy in predicting where they will look.

As more work is conducted in mining software repositories at a large scale, aggregating data from multiple modalities beyond textual representations will prove beneficial. Online video content related to software engineering offers another, thus far largely untapped, venue for source code. In this paper, we present a novel dataset comprised of 111,229 frames extracted from a sample of 100 technology-related YouTube tutorials, taught in the Java and Python programming languages. These videos span a diverse range of Integrated Development Environments (IDEs) and modes of information presentation, lending heterogeneity of image features to our dataset. Hand-labeled for the presence, or lack thereof, of typeset, partially visible typeset, or handwritten code, the dataset provides a uniquely consistent and accurate data source for supervised machine learning ventures in extracting and processing code from images.

The Children's Risk Utility Measure (CRUM) is an interactive, modified version of the Balloon Analogue Risk Task (BART) that is appropriate for young children. This paper utilizes a Bayesian IRT model (an extension of logistic regression) to examine the impact of subject-level characteristics (age, working memory, inhibitory control) and trial-level characteristics (e.g. block, order, etc.) on the performance of children ages 3-6 years on the CRUM. The Bayesian model is especially helpful when performing IRT models-which have many parameters-on small data sets. The order of the trials was not associated with lower difficulty, indicating that children's performance remained steady as the task went on. Working memory performance was negatively associated with children's latent CRUM performance, while high scores on an inhibitory control task were positively related to child's latent CRUM performance. Age and the interaction of age with working memory performance were both negligible effects. The IRT model was an effective way to estimate latent CRUM performance.

Machine learning (ML) techniques can be applied to predict and monitor drought conditions due to climate change. Predicting future vegetation health indicators (such as EVI, NDVI, and LAI) is one approach to forecast drought events for hotspots (e.g. Middle East and North Africa (MENA) regions). Recently, ML models were implemented to predict EVI values using parameters such as land types, time series, historical vegetation indices, land surface temperature, soil moisture, evapotranspiration etc. In this work, we collected the MODIS atmospherically corrected surface spectral reflectance imagery with multiple vegetation related indices for modeling and evaluation of drought conditions in the MENA region. These models are built by a total of 4556 and 519 normalized samples for training and testing purposes, respectively and with 51820 samples used for model evaluation. Models such as multilinear regression, penalized regression models, support vector regression (SVR), neural network, instance-based learning K-nearest neighbor (KNN) and partial least squares were implemented to predict future values of EVI. The models show effective performance in predicting EVI values (R2> 0.95) in the testing and (R2> 0.93) in the evaluation process.

We apply survival analysis methods to a dataset of publicly-available software projects in order to examine the attributes that might lead to their inactivity over time. We ran a Kaplan-Meier analysis and fit a Cox Proportional-Hazards model to a subset of Software Heritage Graph Dataset, consisting of 3052 popular Python projects hosted on GitLab/GitHub, Debian, and PyPI, over a period of 165 months. We show that projects with repositories on multiple hosting services, a timeline of publishing major releases, and a good network of developers, remain healthy over time and should be worthy of the effort put in by developers and contributors.

Weight-sharing is one of the pillars behind Convolutional Neural Networks and their successes. However, in physical neural systems such as the brain, weight-sharing is implausible. This discrepancy raises the fundamental question of whether weight-sharing is necessary. If so, to which degree of precision? If not, what are the alternatives? The goal of this study is to investigate these questions, primarily through simulations where the weight-sharing assumption is relaxed. Taking inspiration from neural circuitry, we explore the use of Free Convolutional Networks and neurons with variable connection patterns. Using Free Convolutional Networks, we show that while weight-sharing is a pragmatic optimization approach, it is not a necessity in computer vision applications. Furthermore, Free Convolutional Networks match the performance observed in standard architectures when trained using properly translated data (akin to video). Under the assumption of translationally augmented data, Free Convolutional Networks learn translationally invariant representations that yield an approximate form of weight-sharing.

Autism Spectrum Disorders are a class of neurodevelopmental disorders which usually present with difficulties in social interactions, verbal and nonverbal forms of communication, repetitive behaviors, and restricted interests. While there are several guides that describe the Americans with Disabilities Act requirements for employers looking to hire those with disabilities, the academic literature describing actual current employer programming to support employees with disabilities is scarce. This study utilizes unsupervised machine learning to explore the policies and practices of 285 employers with regard to autism spectrum disorders, with the ultimate goal of understanding shortcomings in employment practices that can be improved through a combination of educational programs and changes to corporate culture.

Social communication frequently includes nuanced nonverbal communication cues, including eye contact, gestures, facial expressions, body language, and tone of voice. This type of communication is central to face-to-face interaction, but can be challenging for children and adults with autism. Innovative technologies can provide support by augmenting human-delivered cuing and automated prompting. Specifically, immersive virtual reality (VR) offers an option to generalize social skill interventions by concretizing nonverbal information in real-time social interactions. In this work, we explore the design and evaluation of three nonverbal communication applications in immersive VR. The results of this work indicate that delivering real-time visualizations of proximity, speaker volume, and duration of one's speech is feasible in immersive VR and effective for real-time support for proximity regulation for children with autism. We conclude with design considerations for therapeutic VR systems.

Autism Spectrum Disorders (ASD) are a category of developmental disabilities and are categorized by difficulties with social interactions, verbal and nonverbal forms of communication, repetitive behaviors, and restricted interests [1]. Research suggests that young adults with "high functioning" ASD experience significant difficulty in transitioning to work, but little research has examined attitudes, experiences, and needs from the viewpoint of the employer. This study utilized a decision tree to predict likelihood of employment of individuals with "high functioning" ASD based on a survey of 263 representatives from various organizations. The study also analyzes the attributes that are most significant in determining whether an employer will recruit an individual with ASD.

We demonstrate the ability of deep architectures, specifically convolutional neural networks, to learn and differentiate the lexical features of different programming languages presented in coding video tutorials found on the Internet. We analyze over 17,000 video frames containing examples of Java, Python, and other textual and non-textual objects. Our results indicate that not only can computer vision models based on deep architectures be taught to differentiate among programming languages with over 98% accuracy, but can learn language-specific lexical features in the process. This provides a powerful mechanism for carrying out program comprehension research on repositories where source code is represented with imagery rather than text, while simultaneously avoiding the computational overhead of optical character recognition.

While substantial progress has been made in mining code on an Internet scale, efforts to date have been overwhelmingly focused on data sets where source code is represented natively as text. Large volumes of source code available online and embedded in technical videos have remained largely unexplored, due in part to the complexity of extraction when code is represented with images. Existing approaches to code extraction and indexing in this environment rely heavily on computationally intense optical character recognition. To improve the ease and efficiency of identifying this embedded code, as well as identifying similar code examples, we develop a deep learning solution based on convolutional neural networks and autoencoders. Focusing on Java for proof of concept, our technique is able to identify the presence of typeset and handwritten source code in thousands of video images with 85.6%-98.6% accuracy based on syntactic and contextual features learned through deep architectures. When combined with traditional approaches, this provides a more scalable basis for video indexing that can be incorporated into existing software search and mining tools.

We apply cluster analysis to a sample of 2,116 children with Autism Spectrum Disorder in order to identify patterns of challenging behaviors observed in home and centerbased clinical settings. The largest study of this type to date, and the first to employ machine learning, our results indicate that while the presence of multiple challenging behaviors is common, in most cases a dominant behavior emerges. Furthermore, the trend is also observed when we train our cluster models on the male and female samples separately. This work provides a basis for future studies to understand the relationship of challenging behavior profiles to learning outcomes, with the ultimate goal of providing personalized therapeutic interventions with maximum efficacy and minimum time and cost.

We apply a seasonal decomposition time series analysisto TravisTorrent data in order to examine growth trendsand periodic behavior related to number of builds ina continuous integration environment. We apply our techniquesat the macro level using the full TravisTorrent repository consisting of 1,283 projects, and at the micro level considering the Apache Drill project. Our results demonstrate strong seasonal behavior at both the large and small scale using an additive time series model. In addition to being able to accurately capture trend and periodicity in builddata, our techniques are also able to accurately forecast the expected number of builds for a future time interval.

We apply artificial neural networks to the task of predicting the mastery of learning outcomes in response to behavioral therapy for children diagnosed with autism spectrum disorder. We report results for a sample size of 726 children, the largest sample size reported for a study of this nature to date. Our results show that neural networks substantially outperform the linear regression models reported in previous studies, and demonstrate the benefits of leveraging more sophisticated machine learning techniques in the autism research domain.

We analyze naming conventions for classes, interfaces, methods, and fields across 12,151 open-source Java projects. This vocabulary data is then used to train first-order Markov models to classify entity names, as well as to assess adherence to common naming structure. Preliminary results yield an accuracy of 78.34%. Supplementary material may be found at: http://sourcerer.ics.uci.edu/icpc2009/icpc.html.

We adapt latent Dirichlet allocation to the problem of mining bug reports in order to define a new information-theoretic measure of coherence. We then apply our technique to a snapshot of the GNOME Bugzilla database consisting of 431,863 bug reports for multiple software projects. In addition to providing an unsupervised means for modeling report content, our results indicate substantial promise in applying statistical text mining algorithms for estimating bug report quality. Complete results are available from our supplementary materials Web site at http://sourcerer.ics.uci.edu/msr2009/gnome_coherence.html.

The open source movement has made vast quantities of source code available online for free, providing an extremely large dataset for empirical study and potential resuse. A major difficulty in exploiting this potential fully is that the data are currently scattered between competing source code repositories, none of which are structured for empirical analysis and cross-project comparison. As a result, software researchers and developers are left to compile their own datasets, resulting in duplicated effort and limited results. To address this challenge, we built SourcererDB, an aggregated repository of statically analyzed and cross-linked open source Java projects. SourcererDB contains local snapshots of 2,852 Java projects taken from Sourceforge, Apache and Java.net. These projects are statically analyzed to extract rich structural information, which is then stored in a relational database. References to entities in the 16,058 external jars are resolved and grouped, allowing for cross-project usage information to be accessed easily. This paper describes: (a) the mechanism for resolving and grouping these cross-project references, (b) the structure of and the metamodel for the SourcererDB repository, and (d) end-user dataset access mechanisms. Our goal in building SourcererDB is to provide a rich dataset of source code to facilitate the sharing of extracted data and to encourage reuse and repeatability of experiments.

We develop and apply unsupervised statistical topic models, in particular latent Dirichlet allocation, to identify functional components of source code and study their evolution over multiple project versions. We present results for two large, open source Java projects, Eclipse and Argo UML, which are well-known and well-studied within the software mining community. Our results demonstrate the effectiveness of probabilistic topic models in automatically summarizing the temporal dynamics of software concerns, with direct application to project management and program understanding. In addition to detecting the emergence of topics on the release timeline which represent integration points for key source code functionality, our techniques can also be used to pinpoint refactoring events in the underlying software design, as well as to identify general programming concepts whose prevalence is dependent only of the size of the code base to be analyzed. Complete results are available from our supplementary materials website at http://sourcerer.ics.uci.edu/icmla2008/software_evolution.html.

After more than 10 years, Aspect-Oriented Programming (AOP) is still a controversial idea. While the concept of aspects appeals to everyone's intuitions, concrete AOP solutions often fail to convince researchers and practitioners alike. This discrepancy results in part from a lack of an adequate theory of aspects, which in turn leads to the development of AOP solutions that are useful in limited situations.

We develop and apply statistical topic models to software as a means of extracting concepts from source code. The effectiveness of the technique is demonstrated on 1,555 projects from SourceForge and Apache consisting of 113,000 files and 19 million lines of code. In addition to providing an automated, unsupervised, solution to the problem of summarizing program functionality, the approach provides a probabilistic framework with which to analyze and visualize source file similarity. Finally, we introduce an information-theoretic approach for computing tangling and scattering of extracted concepts, and present preliminary results

Large repositories of source code create new challenges and opportunities for statistical machine learning. Here we first develop an infrastructure for the automated crawling, parsing, and database storage of open source software. The infrastructure allows us to gather Internet-scale source code. For instance, in one experiment, we gather 4,632 java projects from SourceForge and Apache totaling over 38 million lines of code from 9,250 developers. Simple statistical analyses of the data first reveal robust power-law behavior for package, SLOC, and method call distributions. We then develop and apply unsupervised author-topic, probabilistic models to automatically discover the topics embedded in the code and extract topic-word and author-topic distributions. In addition to serving as a convenient summary for program function and developer activities, these and other related distributions provide a statistical and information-theoretic basis for quantifying and analyzing developer similarity and competence, topic scattering, and document tangling, with direct applications to software engineering. Finally, by combining software textual content with structural information captured by our CodeRank approach, we are able to significantly improve software retrieval performance, increasing the AUC metric to 0.86-- roughly 10-30% better than previous approaches based on text alone.

UNLABELLED: ChemDB is a chemical database containing nearly 5M commercially available small molecules, important for use as synthetic building blocks, probes in systems biology and as leads for the discovery of drugs and other useful compounds. The data is publicly available over the web for download and for targeted searches using a variety of powerful methods. The chemical data includes predicted or experimentally determined physicochemical properties, such as 3D structure, melting temperature and solubility. Recent developments include optimization of chemical structure (and substructure) retrieval algorithms, enabling full database searches in less than a second. A text-based search engine allows efficient searching of compounds based on over 65M annotations from over 150 vendors. When searching for chemicals by name, fuzzy text matching capabilities yield productive results even when the correct spelling of a chemical name is unknown, taking advantage of both systematic and common names. Finally, built in reaction models enable searches through virtual chemical space, consisting of hypothetical products readily synthesizable from the building blocks in ChemDB. AVAILABILITY: ChemDB and Supplementary Materials are available at http://cdb.ics.uci.edu. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.