The emergence of a new paradigm in machine learning known as semi-supervised learning (SSL) has seen benefits to many applications where labeled data is expensive to obtain. However, unlike supervised learning (SL), which enjoys a rich and deep theoretical foundation, semi-supervised learning, which uses additional unlabeled data for training, still remains a theoretical mystery lacking a sound fundamental understanding. The purpose of this research thesis is to take a first step towards bridging this theory-practice gap. We focus on investigating the inherent limitations of the benefits SSL can provide over SL. We develop a framework under which one can analyze the potential benefits, as measured by the sample complexity of SSL. Our framework is utopian in the sense that a SSL algorithm trains on a labeled sample and an unlabeled distribution, as opposed to an unlabeled sample in the usual SSL model. Thus, any lower bound on the sample complexity of SSL in this model implies lower bounds in the usual model. Roughly, our conclusion is that unless the learner is absolutely certain there is some non-trivial relationship between labels and the unlabeled distribution (``SSL type assumption''), SSL cannot provide significant advantages over SL. Technically speaking, we show that the sample complexity of SSL is no more than a constant factor better than SL for any unlabeled distribution, under a no-prior-knowledge setting (i.e. without SSL type assumptions). We prove that for the class of thresholds in the realizable setting the sample complexity of SL is at most twice that of SSL. Also, we prove that in the agnostic setting for the classes of thresholds and union of intervals the sample complexity of SL is at most a constant factor larger than that of SSL. We conjecture this to be a general phenomenon applying to any hypothesis class. We also discuss issues regarding SSL type assumptions, and in particular the popular cluster assumption. We give examples that show even in the most accommodating circumstances, learning under the cluster assumption can be hazardous and lead to prediction performance much worse than simply ignoring the unlabeled data and doing supervised learning. We conclude with a look into future research directions that build on our investigation.

This thesis studies two problems in theoretical machine learning. The first part of the thesis investigates the statistical stability of clustering algorithms. In the second part, we study the relative advantage of having unlabeled data in classification problems. Clustering stability was proposed and used as a model selection method in clustering tasks. The main idea of the method is that from a given data set two independent samples are taken. Each sample individually is clustered with the same clustering algorithm, with the same setting of its parameters. If the two resulting clusterings turn out to be close in some metric, it is concluded that the clustering algorithm and the setting of its parameters match the data set, and that clusterings obtained are meaningful. We study asymptotic properties of this method for certain types of cost minimizing clustering algorithms and relate their asymptotic stability to the number of optimal solutions of the underlying optimization problem. In classification problems, it is often expensive to obtain labeled data, but on the other hand, unlabeled data are often plentiful and cheap. We study how the access to unlabeled data can decrease the amount of labeled data needed in the worst-case sense. We propose an extension of the probably approximately correct (PAC) model in which this question can be naturally studied. We show that for certain basic tasks the access to unlabeled data might, at best, halve the amount of labeled data needed.

<p>Secession is a watershed event not only for the new state that is created and the old state that is dissolved, but also for neighboring states, proximate ethno-political groups and major powers. This project examines the problem of violent secessionist conflict and addresses an important debate at the intersection of comparative and international politics about the conditions under which secession is a peaceful solution to ethnic conflict. It demonstrates that secession is rarely a solution to ethnic conflict, does not assure the protection of remaining minorities and produces new forms of violence. To explain why some secessions produce peace, while others generate violence, the project develops a theoretical model of the conditions that produce internally coherent, stable and peaceful post-secessionist states rather than recursive secession (i.e., secession from a new secessionist state) or interstate disputes between the rump and secessionist state. Theoretically, the analysis reveals a curvilinear relationship between ethno-territorial heterogeneity and conflict, explains disparate findings in the literature on ethnic conflict and conclusively links ethnic structure and violence. The project also contributes to the literature on secessionist violence, and civil war more generally, by linking intrastate and interstate causes, showing that what is frequently thought of as a domestic phenomenon is in fact mostly a phenomenon of international politics. Drawing upon original data, methodological advances at the interface of statistics, computer science and probability theory, and qualitative methods such as elite interviews and archival research, the project offers a comprehensive, comparative and contextual treatment of secession and violence.</p> / Dissertation

This thesis presents an unsupervised method for discovering and analyzing the different kinds of activities in an active environment. Drawing from natural language processing, a novel representation of activities as bags of event n-grams is introduced, where the global structural information of activities using their local event statistics is analyzed. It is demonstrated how maximal cliques in an undirected edge-weighted graph of activities, can be used in an unsupervised manner, to discover the different activity-classes. Taking on some work done in computer networks and bio-informatics, it is shown how to characterize these discovered activity-classes from a wholestic as well as a by-parts view-point. A definition of anomalous activities is formulated along with a way to detect them based on the difference of an activity instance from each of the discovered activity-classes. Finally, an information theoretic method to explain the detected anomalies in a human-interpretable form is presented. Results over extensive data-sets, collected from multiple active environments are presented, to show the competence and generalizability of the proposed framework.

<p>Vi gör dagligen bedömningar av risker i situationer av alla de slag. Trafiken är en av de situationer där riskbedömningar är särskilt viktiga och hur korrekt bedömningen blir beror på hur vi upplever riskerna, vilket i hög grad varierar från person till person. Hypotesen i detta examensarbete är: Kvinnliga bilister upplever större subjektiv risk i trafiken än manliga bilister. Undersökningen av hypotesens giltighet utfördes med hjälp av intervjuer av 20 män och kvinnor. Frågorna syftade till att ta reda på deltagarnas subjektiva upplevelser av trafikrisker. Resultaten pekade på att det finns en tendens till skillnad i hur de båda könen upplever risker i trafiken. Kunskap om hur kvinnor och män upplever risker i trafiken är användbara för att göra olika trafiksituationer säkrare och därmed minska antalet olyckor. Ännu ett steg i denna strävan är att undersöka hur stressfaktorer påverkar riskuppfattningen i en trafiksituation.</p>

<p>Higher education is undergoing continuous changes and new learning tools and methods are implemented. Researchers in education do not always agree upon the effectiveness of some of the methods introduced into engineering education. The present thesis consists of two case studies on educational methods introduced at the Department of Energy Technology, at Royal Institute of Technology (KTH), Sweden. The qualitative research methodology has been used in case one and a combination of qualitative and quantitative methodology has been used in the second case. The sources of evidences consisted of: unstructured interviews, analysis of video recording, questionnaires, and analysis of a variety of documents. In the first case, an educational program in heat and power technology was analysed. The second case consists in an in-depth study of group dynamics in a Problem –Based Learning course. These studies showed that the learning approach adopted by students depends strongly on the way they view the particular learning tool or method. The first case study revealed the existence of two types of learners. Surfacelearners follow the structure suggested by the designers of the multimedia program. This category of learners focuses only on the material available in the program. Deep-learners go beyond the information and the structure suggested in the program and combine different learning tools in their learning. These students do not follow the structure of the tutorials’ of the multimedia program. This study showed that students who had a strong view how to learn with a multimedia program or a learning method benefited less from the learning tools available. Students with weak views on how to learn from educational program or leaning tool benefit less from the presentation and engage in more surface learning. Self-motivated learners use the multimedia presentation in novel ways and crosscheck the information given with other material. The second study showed that students have unclear and weak views on how to learn with student-directed Problem- Based Learning model. Four types of learners were identified in Problem-Based Learning project: Leaders, Key Actors, Common Students and Social Loafers. Leaders and Key Actors are self-motivated individuals and participate most in the projects. Students who viewed themselves or were viewed as leaders were held responsible to take most of the decisions and students expected them to work more than the average student. Students who viewed themselves as common team members expected a lower workload than leaders’. Key Actors are self-motivated students who do not view themselves as separate from other group members but who participate more than others. Leaders learned more group and social processes, that they did not fully take part in, while common students learned more from the project management aspects that they did not take part in. The study also found that Problem-Based Learning groups can become very cohesive, and can develop distorted views on how to learn with Problem-Based Learning, and un-common group dynamics phenomena such as groupthink can occur in Problem-Based Learning setting.</p>

Title from title page of PDF (University of Missouri--St. Louis, viewed February 24, 2010). Includes bibliographical references (p. 29-30).

This essay discusses if teaching English literature in high school classes, in accordance with the socio-cultural learning theory, can be considered to promote language learning substantially better than teaching English literature in accordance with the transmission theory. This essay also investigates and compares how well teaching English literature, in accordance with each of these two learning theories, fulfills stipulations in the Swedish National Curriculum for high school courses English 5, 6 and 7. In order to show differences between the socio-cultural learning theory and the transmission theory there are presentations and discussions of different teaching strategies and learning tasks/exercises in accordance with each of these two learning theories. The base for argumentation in this essay is constituted by analytical and theoretical studies of teaching English literature in accordance with the socio-cultural learning theory and in accordance with the transmission theory. There are also theoretical studies of the Swedish National Curriculum (of high school courses English 5, 6 and 7) and previous empirical research and studies (which include teaching and/or language learning and the socio-cultural learning theory). This essay also includes a presentation and discussion of advantages and disadvantages for each learning theory.

Programmed Instruction was a dominant force in educational research during the 1960's and early 1970's. During this period a new cognitively oriented psychology of education arose that eventually replaced the behaviorally driven research into programmed instruction. For some reason little work was done on integrating these two approaches. This study was designed to test a programmed instructional system based on the intrinsic programmed instructional technology of Crowder (1961) but incorporating design elements derived from cognitive load theory (Sweller, 1988). Specifically, this study tested a multimedia programmed instructional system (CLT-PI) based on work by Mayer (2001) against a text based system (TXT), a traditional linear programmed instructional system based on Skinner (1954, 1958)(LPI), and an intrinsic programmed instructional system based on the work of Crowder (1960, 1961)(IPI). Three hypotheses were tested; 1) Programmed instruction would produce higher long-term retention and transfer scores than TXT, 2) IPI would produced higher long-term retention and transfer scores than LPI, and 3) CLT-PI would produce higher long-term retention and transfer scores than LPI or IPI. 115 undergraduate university students were randomly assigned to one of four conditions where they studied a chapter on human sleep. Each condition was presented on a computer with a test following study. A long-term test was given 4 weeks later. Both short- and long-term tests contained retention and transfer questions. Analyses were conducted using repeated measures MANOVA. A series of ANOVA tests were conducted to determine specific effects and interactions. The first hypothesis was partly supported in that CLT-PI and IPI produced higher long-term retention scores than TXT. LPI, however, did not.

The primarily goal of this study was to examine the patterns of cognitive and language processing of children with Specific Language Impairment (SLI) and how they related to their text comprehension, interpreted within constructivism. The study aimed to characterize the difference in language, character decoding, metacognitive processing and text comprehension between children with SLI and those under typical development; to identify the inter-relationships among their language processing, character decoding, metacognitive processing and text comprehension; and to investigate how the implementation of cognitive reading strategies instruction change their language processing, character decoding, metacognitive processing, belief towards reading and text comprehension. Two studies were carried out. In Study One, 73 participants were recruited from two Hong Kong primary schools; they were at second and third grade, 42 were diagnosed of SLI and the other were under typical development. Standardized instruments were used to tap children’s language processing and character decoding respectively. Researcher developed Metacognitive Processing Scale were adopted to rate their metacognitive and deep processing of text. A set of comprehension test, comprised of forced-choice inferential questions and two recall tasks, were used to assess their depth in understanding different types of text. Analyses indicated that SLI students did not only score poorer in language processing, but also in character decoding, metacognitive processing and text comprehension. Further analyses of both the entire sample and the SLI sample, indicated that there were significant correlations between character decoding, language processing, metacognitive processing with children’s text comprehension scores. In Study Two, there were 40 participants recruited from the SLI sample of the Study One. Cognitive reading strategy instruction program were developed. 21 participants was randomly selected to receive the experimental instruction and the another 19 were under conventional instruction as control. Besides the measures used in the Study One, interviews and teacher-reporting questionnaires were used to tap children’s belief towards reading and their classroom engagement. Analysis of pre-instruction and post-instruction tests indicated the experimental children showed significantly better progress on their oral language, text comprehension and belief about reading. Both the experimental and the control group showed similar progress on character decoding. The study offers both theoretical and educational contribution on the literacy development among the population of SLI. It identifies the role of metacognitive processing on literacy achievement. It provides the evidence of implementing cognitively-based reading strategies for literacy instruction for children with SLI within Chinese context. Upon the introduction of inclusive education, teachers now face students with much wider diversity, including a significant number of children with SLI. Possible collaboration between frontline teachers and speech therapists in designing potential classroom activities is discussed. / published_or_final_version / Education / Doctoral / Doctor of Education