Thesis (Ph. D.)--Oregon State University, 2000. / Typescript (photocopy). Includes bibliographical references (leaves 282-294). Also available on the World Wide Web.

Thesis (Ed.D.) -- Central Connecticut State University, 2005. / Thesis advisor: Richard Arends. "... in partial fulfillment of the requirements for the degree of Doctor of Education." Includes bibliographical references (leaves 61-72). Also available via the World Wide Web.

Intelligent tutoring systems are effective for improving students’ learning outcomes (Bowen et al., 2013; Koedinger & Anderson, 1997; Pane et al., 2013). However, constructing tutoring systems that are pedagogically effective has been widely recognized as a challenging problem (Murray, 1999, 2003). In this thesis, I explore the use of computational models of apprentice learning, or computer models that learn interactively from examples and feedback, to support tutor development. In particular, I investigate their use for authoring expert-models via demonstrations and feedback (Matsuda et al., 2014), predicting student behavior within tutors (VanLehn et al., 1994), and for testing alternative learning theories (MacLellan, Harpstead, Patel, & Koedinger, 2016). To support these investigations, I present the Apprentice Learner Architecture, which posits the types of knowledge, performance, and learning components needed for apprentice learning and enables the generation and testing of alternative models. I use this architecture to create two models: the DECISION TREE model, which non- incrementally learns when to apply its skills, and the TRESTLE model, which instead learns incrementally. Both models both draw on the same small set of prior knowledge for all simulations (six operators and three types of relational knowledge). Despite their limited prior knowledge, I demonstrate their use for efficiently authoring a novel experimental design tutor and show that they are capable of achieving human-level performance in seven additional tutoring systems that teach a wide range of knowledge types (associations, categories, and skills) across multiple domains (language, math, engineering, and science). I show that the models are capable of predicting which versions of a fraction arithmetic and box and arrows tutors are more effective for human students’ learning. Further, I use a mixedeffects regression analysis to evaluate the fit of the models to the available human data and show that across all seven domains the TRESTLE model better fits the human data than the DECISION TREE model, supporting the theory that humans learn the conditions under which skills apply incrementally, rather than non-incrementally as prior work has suggested (Li, 2013; Matsuda et al., 2009). This work lays the foundation for the development of a Model Human Learner— similar to Card, Moran, and Newell’s (1986) Model Human Processor—that encapsulates psychological and learning science findings in a format that researchers and instructional designers can use to create effective tutoring systems.

Key to achieving more effective machine intelligence is the capability to generalise knowledge across different contexts. In this thesis, we develop a new and very general perspective on knowledge sharing that unifi es and generalises many existing methodologies, while being practically effective, simple to implement, and opening up new problem settings. Knowledge sharing across tasks and domains has conventionally been studied disparately. We fi rst introduce the concept of a semantic descriptor and a flexible neural network approach to knowledge sharing that together unify multi-task/multi-domain learning, and encompass various classic and recent multi-domain learning (MDL) and multi-task learning (MTL) algorithms as special cases. We next generalise this framework from single-output to multi-output problems and from shallow to deep models. To achieve this, we establish the equivalence between classic tensor decomposition methods, and specifi c neural network architectures. This makes it possible to implement our framework within modern deep learning stacks. We present both explicit low-rank, and trace norm regularisation solutions. From a practical perspective, we also explore a new problem setting of zero-shot domain adaptation (ZSDA) where a model can be calibrated solely based on some abstract information of a new domain, e.g., some metadata like the capture device of photos, without collecting or labelling the data.

In this thesis we are designing machine learning methodologies for solving the problem of video annotation and retrieval using either pre-defined semantic concepts or ad-hoc queries. Concept-based video annotation refers to the annotation of video fragments with one or more semantic concepts (e.g. hand, sky, running), chosen from a predefined concept list. Ad-hoc queries refer to textual descriptions that may contain objects, activities, locations etc., and combinations of the former. Our contributions are: i) A thorough analysis on extending and using different local descriptors towards improved concept-based video annotation and a stacking architecture that uses in the first layer, concept classifiers trained on local descriptors and improves their prediction accuracy by implicitly capturing concept relations, in the last layer of the stack. ii) A cascade architecture that orders and combines many classifiers, trained on different visual descriptors, for the same concept. iii) A deep learning architecture that exploits concept relations at two different levels. At the first level, we build on ideas from multi-task learning, and propose an approach to learn concept-specific representations that are sparse, linear combinations of representations of latent concepts. At a second level, we build on ideas from structured output learning, and propose the introduction, at training time, of a new cost term that explicitly models the correlations between the concepts. By doing so, we explicitly model the structure in the output space (i.e., the concept labels). iv) A fully-automatic ad-hoc video search architecture that combines concept-based video annotation and textual query analysis, and transforms concept-based keyframe and query representations into a common semantic embedding space. Our architectures have been extensively evaluated on the TRECVID SIN 2013, the TRECVID AVS 2016, and other large-scale datasets presenting their effectiveness compared to other similar approaches.

In this thesis we study deep learning architectures for the problem of human action recognition in image sequences, i.e. the problem of automatically recognizing what people are doing in a given video. As unlabeled video data is easily accessible these days, we first explore models that can learn meaningful representations of sequences without actually having to know what is happening in the sequences at hand. More specifically, we first explore the convolutional restricted Boltzmann machine (RBM) and show how a stack of convolutional RBMs can be used to learn and extract features from sequences in an unsupervised way. Using the classical Fisher vector pipeline to encode the extracted features we apply them on the task of action classification. We move on to feature extraction using larger, deep convolutional neural networks and propose a novel architecture which expresses the processing steps of the classical Fisher vector pipeline as network layers. By contrast to other methods where these steps are performed consecutively and the corresponding parameters are learned in an unsupervised manner, defining them as a single neural network allows us to refine the whole model discriminatively in an end to end fashion. We show that our method achieves significant improvements in comparison to the classical Fisher vector extraction chain and results in a comparable performance to other convolutional networks, while largely reducing the number of required trainable parameters. Finally, we explore how the proposed architecture can be modified into a hybrid network that combines the benefits of both unsupervised and supervised training methods, resulting in a model that learns a semi-supervised Fisher vector descriptor of the input data. We evaluate the proposed model at image classification and action recognition problems and show how the model's classification performance improves as the amount of unlabeled data increases during training.

Demand response (DR) is one of the integral mechanisms of today's smart grids. It enables consumer energy assets such as flexible loads, standby generators and storage systems to add value to the grid by providing cost-effective flexibility. With increasing renewable generation and impending electric vehicle deployment, there is a critical need for large volumes of reliable and responsive flexibility through DR. This poses a new challenge for the electricity sector. Smart grid development has resulted in the availability of large amounts of data from different physical segments of the grid such as generation, transmission, distribution and consumption. For instance, smart meter data carrying valuable information is increasingly available from the consumers. Parallel to this, the domain of data analytics and machine learning (ML) is making immense progress. Data-driven modelling based on ML algorithms offers new opportunities to utilise the smart grid data and address the DR challenge. The thesis demonstrates the use of data-driven models for enhancing DR from large consumers such as commercial and industrial (C&I) buildings. A reliable, computationally efficient, cost-effective and deployable data-driven model is developed for large consumer building load estimation. The selection of data pre-processing and model development methods are guided by these design criteria. Based on this model, DR operational tasks such as capacity scheduling, performance evaluation and reliable operation are demonstrated for consumer energy assets such as flexible loads, standby generators and storage systems. Case studies are designed based on the frameworks of ongoing DR programs in different electricity markets. In these contexts, data-driven modelling shows substantial improvement over the conventional models and promises more automation in DR operations. The thesis also conceptualises an emissions-based DR program based on emissions intensity data and consumer load flexibility to demonstrate the use of smart grid data in encouraging renewable energy consumption. Going forward, the thesis advocates data-informed thinking for utilising smart grid data towards solving problems faced by the electricity sector.

This study explored the classroom learning environment in science among kindergarten students. In particular, I investigated both students' and their parents' perceptions of both preferred and actual learning environments. Additionally, I explored associations between student outcomes (achievement and attitudes toward science) and the nature of the classroom learning environment (as perceived by students and by their parents). The study involved the construction and validation of a learning environment questionnaire that was used by both parents and kindergarten students. Although the questionnaire was validated for use with five- and six-year-old kindergarten students, the same format was used for both parents and students. Prior learning environment studies (Fraser, 1998a) typically have involved the use of questionnaires neither by parents (with a notable exception being the recent study by Allen and Fraser, 2002) or by such young students. There is little doubt that, in just two decades, the field of classroom learning environment has progressed enormously (Fraser, 1998a) and that research involving qualitative methods and research involving quantitative methods each have made outstanding contributions to this overall progress (Tobin & Fraser, 1998). A historical look at the field of learning environments over the past few decades shows that a striking feature is the availability of a variety of economical, valid and widely applicable questionnaires for assessing student perceptions of classroom environments (Fraser, 1998b). This learning environment study is significant not only because it involves very young students (kindergarten) and their parents, but also a classroom learning environment questionnaire was developed and validated in Spanish, for both students and parents. / The design of the study involved a sample of 172 kindergarteners from six classes and 78 parents of the same students from the same six classes. The ethnic make-up for this group of 172 students was 11.8% White, 49% Black, 33.6% Hispanic, and 5.6% of other nationalities. The gender breakdown was 40.4% boys and 59.6% girls. Approximately 45% of the kindergarten student population was made up of English Speakers of Other Languages (ESOL) students. The instruments used included modified versions in English and Spanish of the What Is Happening In This Class (WIHIC)? questionnaire and of the Test of Science-Related Attitudes (TOSRA). A major finding of the study was that the modified version of the What Is Happening In This Class? (WIHIC) questionnaire in the English and Spanish languages displayed satisfactory factorial validity and internal consistency reliability when used with kindergarten students and their parents. Secondly, parents perceived a more favorable actual classroom environment than did kindergarten students, but students preferred a much more favorable classroom environment than did their parents. The magnitudes of differences between students and parents are greater for the preferred form than the actual form. Finally, statistically significant associations were found between kindergarten students' perceptions of the. classroom environment and the outcomes of achievement and attitudes to science.

Thesis (Ph. D.)--Rutgers University, 2009. / "Graduate Program in Computer Science." Includes bibliographical references (p. 238-261).

Thesis (M.S.)--Rochester Institute of Technology, 1989. / Includes bibliographical references (leaves 96-97).