In this thesis, we focus on building reliable hybrid human-machine classifiers to be deployed in cost-sensitive classification tasks. The objective is to assess ML quality in hybrid classification contexts and design the appropriate metrics, thereby knowing whether we can trust the model predictions and identifying the subset of items on which the model is well-calibrated and trustworthy. We start by discussing the key concepts, research questions, challenges, and architecture to design and implement an effective hybrid classification service. We then present a deeper investigation of each service component along with our solutions and results. We mainly contribute to cost-sensitive hybrid classification, selective classification, model calibration, and active learning. We highlight the importance of model calibration in hybrid classification services and propose novel approaches to improve the calibration of human-machine classifiers. In addition, we argue that the current accuracy-based metrics are misaligned with the actual value of machine learning models and propose a novel metric ``value". We further test the performance of SOTA machine learning models in NLP tasks with a cost-sensitive hybrid classification context. We show that the performance of the SOTA models in cost-sensitive tasks significantly drops when we evaluate them according to value rather than accuracy. Finally, we investigate the quality of hybrid classifiers in the active learning scenarios. We review the existing active learning strategies, evaluate their effectiveness, and propose a novel value-aware active learning strategy to improve the performance of selective classifiers in the active learning of cost-sensitive tasks.

No description available.

No description available.

Traditional multi-class classification problems assume that each instance is associated with a single label from category set Y where |Y| > 2. Multi-label classification generalizes multi-class classification by allowing each instance to be associated with multiple labels from Y. In many real world data analysis problems, data objects can be assigned into multiple categories and hence produce multi-label classification problems. For example, an image for object categorization can be labeled as 'desk' and 'chair' simultaneously if it contains both objects. A news article talking about the effect of Olympic games on tourism industry might belong to multiple categories such as 'sports', 'economy', and 'travel', since it may cover multiple topics. Regardless of the approach used, multi-label learning in general requires a sufficient amount of labeled data to recover high quality classification models. However due to the label sparsity, i.e. each instance only carries a small number of labels among the label set Y, it is difficult to prepare sufficient well-labeled data for each class. Many approaches have been developed in the literature to overcome such challenge by exploiting label correlation or label dependency. In this dissertation, we propose a probabilistic model to capture the pairwise interaction between labels so as to alleviate the label sparsity. Besides of the traditional setting that assumes training data is fully labeled, we also study multi-label learning under other scenarios. For instance, training data can be unreliable due to missing values. A conditional Restricted Boltzmann Machine (CRBM) is proposed to take care of such challenge. Furthermore, labeled training data can be very scarce due to the cost of labeling but unlabeled data are redundant. We proposed two novel multi-label learning algorithms under active setting to relieve the pain, one for standard single level problem and one for hierarchical problem. Our empirical results on multiple multi-label data sets demonstrate the efficacy of the proposed methods. / Computer and Information Science

Depth sensing is pivotal in robotics; however, monocular depth estimation encounters significant challenges. Existing algorithms relying on large-scale labeled data and large Deep Convolutional Neural Networks (DCNNs) hinder real-world applications. We propose two lightweight architectures that achieve commendable accuracy rates of 91.2% and 90.1%, simultaneously reducing the Root Mean Square Error (RMSE) of depth to 4.815 and 5.036. Our lightweight depth model operates at 29-44 FPS on the Jetson Nano GPU, showcasing efficient performance with minimal power consumption. Moreover, we introduce a mask network designed to visualize and analyze the compact depth network, aiding in discerning informative samples for the active learning approach. This contributes to increased model accuracy and enhanced generalization capabilities. Furthermore, our methodology encompasses the introduction of an active learning framework strategically designed to enhance model performance and accuracy by efficiently utilizing limited labeled training data. This novel framework outperforms previous studies by achieving commendable results with only 18.3% utilization of the KITTI Odometry dataset. This performance reflects a skillful balance between computational efficiency and accuracy, tailored for low-cost devices while reducing data training requirements.:1. Introduction 2. Literature Review 3. AI Technologies for Edge Computing 4. Monocular Depth Estimation Methodology 5. Implementation 6. Result and Evaluation 7. Conclusion and Future Scope Appendix

This thesis explores the integration of digital technologies and active learning methodologies in physics education, focusing on both high school and introductory undergraduate laboratory courses. The research is motivated by the need to move away from traditional, teacher-centered approaches and embrace methods that actively engage students in the learning process. The first part of the thesis details the implementation of the E-CLASS (Colorado Learning Attitudes about Science Survey for Experimental Physics) survey in Italian undergraduate courses. This survey provided insights into students' attitudes towards experimental physics, guiding curriculum refinement to enhance learning outcomes. By analysing pre- and post-course data, we identified areas for improvement and adjusted teaching practices accordingly. The second part of the research focuses on the introduction of Jupyter Notebooks with Python in laboratory courses. We defined and introduced a set of laboratory computational learning goals that were incorporated into the course to foster students' abilities to write Python codes for data manipulation, analysis, and visualization, as well as to effectively communicate their work using Jupyter Notebooks. This approach aimed to lower the entry barrier to programming, enhancing students' computational skills, which are fundamental for modern scientific methodologies, and self-efficacy in a manner more aligned with professional physics practices. The effectiveness of the approach is described on the basis of multi-step assessments aligned with the defined learning goals. The third part of this research focuses on the implementation of the Investigative Science Learning Environment (ISLE) approach in high school physics courses using iOLab devices. ISLE is an inquiry-based methodology where students learn physics by practicing it, mirroring the activities of professional physicists. This approach involves students working in groups in generating and testing their own explanations for observed phenomena through hands-on experimentation. The ISLE methodology fosters critical thinking, problem-solving, and collaboration, essential for learning scientific practices. Initial results from the implementation show that students were highly engaged and appreciated the use of technology and group work in their learning process, although longer interventions are needed to significantly impact students' habits. In addition, a teaching module on introducing the FFT spectrum as a graphical representation to explore sound phenomena was presented using Jupyter Notebooks and smartphone sensors, further integrating computational elements into the curriculum. In conclusion, this thesis shows the potential of digital technologies and active learning methodologies in improving student learning. By fostering critical thinking, data analysis skills and scientific inquiry, these approaches significantly enhance the educational experience and prepare students for the complexities of 21st- century world.

Yes / A recent pilot study by Chesham et al. in BMC Medicine established some initial effects of the Daily Mile™ using a quasi-experimental repeated measures design, with valid and reliable outcome assessments for moderate-to-vigorous physical activity, fitness and body composition. Their contribution is important and welcome, yet, alone, it is insufficient to justify the recent UK-wide adoption of the Daily Mile within the Childhood Obesity Plan. The study concluded that the Daily Mile had positive effects on moderate-to-vigorous physical activity, fitness and body composition, suggesting that intervention effectiveness was confirmed. However, only some of the significant limitations of the work were addressed. Herein, we identify and discuss six key limitations, which, combined, suggest a more tentative conclusion. In summary, evidence supporting the effectiveness of the Daily Mile is in its infancy and requires refinement to fully justify its widespread adoption. Further, we need to be cautious considering that the full range of its impacts, both positive and negative, remain to be fully established.

Yes / Physically active learning (PAL) - integration of movement within delivery of academic content - is a core component of many whole-of-school physical activity approaches. Yet, PAL intervention methods and strategies vary and frequently are not sustained beyond formal programmes. To improve PAL training, a more comprehensive understanding of the behavioural and psychological processes that influence teachers' adoption and implementation of PAL is required. To address this, we conducted a meta-synthesis to synthesise key stakeholders' knowledge of facilitators and barriers to teachers' implementing PAL in schools to improve teacher-focussed PAL interventions in primary (elementary) schools. We conducted a meta-synthesis using a five-stage thematic synthesis approach to; develop a research purpose and aim, identify relevant articles, appraise studies for quality, develop descriptive themes and interpret and synthesise the literature. In the final stage, 14 domains from the Theoretical Domain Framework (TDF) were then aligned to the final analytical themes and subthemes. We identified seven themes and 31 sub-themes from 25 eligible papers. Four themes summarised teacher-level factors: PAL benefits, teachers' beliefs about own capabilities, PAL teacher training, PAL delivery. One theme encompassed teacher and school-level factors: resources. Two themes reflected school and external factors that influence teachers' PAL behaviour: whole-school approach, external factors. Ten (of 14) TDF domains aligned with main themes and sub-themes: Knowledge, Skills, Social/Professional Role and Identity, Beliefs about Capabilities, Beliefs about Consequences, Reinforcement, Goals, Environmental Context and Resources, Social influences and Emotion. Our synthesis illustrates the inherent complexity required to change and sustain teachers' PAL behaviours. Initially, teachers must receive the training, resources and support to develop the capability to implement and adapt PAL. The PAL training programme should progress as teachers' build their experience and capability; content should be 'refreshed' and become more challenging over time. Subsequently, it is imperative to engage all levels of the school community for PAL to be fully integrated into a broader school system. Adequate resources, strong leadership and governance, an engaged activated community and political will are necessary to achieve this, and may not currently exist in most schools. / European Union ERASMUS+ Strategic Partnership Fund as part of the Activating Classroom Teachers (ACTivate)- teachers on the move project (NO: 2019–1-NO01-KA203–060324)

Achieving robust and reliable autonomous operation even in complex unstructured environments is a central goal of field robotics. As the environments and scenarios to which robots are applied have continued to grow in complexity, so has the challenge of properly defining preferences and tradeoffs between various actions and the terrains they result in traversing. These definitions and parameters encode the desired behavior of the robot; therefore their correctness is of the utmost importance. Current manual approaches to creating and adjusting these preference models and cost functions have proven to be incredibly tedious and time-consuming, while typically not producing optimal results except in the simplest of circumstances. This thesis presents the development and application of machine learning techniques that automate the construction and tuning of preference models within complex mobile robotic systems. Utilizing the framework of inverse optimal control, expert examples of robot behavior can be used to construct models that generalize demonstrated preferences and reproduce similar behavior. Novel learning from demonstration approaches are developed that offer the possibility of significantly reducing the amount of human interaction necessary to tune a system, while also improving its final performance. Techniques to account for the inevitability of noisy and imperfect demonstration are presented, along with additional methods for improving the efficiency of expert demonstration and feedback. The effectiveness of these approaches is confirmed through application to several real world domains, such as the interpretation of static and dynamic perceptual data in unstructured environments and the learning of human driving styles and maneuver preferences. Extensive testing and experimentation both in simulation and in the field with multiple mobile robotic systems provides empirical confirmation of superior autonomous performance, with less expert interaction and no hand tuning. These experiments validate the potential applicability of the developed algorithms to a large variety of future mobile robotic systems.

Thesis (PhD)--Stellenbosch University, 2014. / ENGLISH ABSTRACT: This study is a multi-site ethnography that focuses on young people from one low-income, Cape Town neighbourhood, an area that I got to know well between 2008 and 2012, when I worked and conducted research there. I explore how young people from this area, that I call Rosemary Gardens, learn in three different places. These places are, firstly, classrooms at Rosemary Gardens High School, secondly, a community-based hip-hop/ rap group called the Doodvenootskap, and, thirdly, a youth radio show called Youth Amplified, which involved many young people from Rosemary Gardens. In each of the three places a ‘spatio-dialogical’ analysis was used to examine learning that emerges through collaborative interactions between people. Dialogic learning may take place when young people are exposed to multiple, different perspectives, which manifest through language. This form of learning is ‘spatialised’ because it occurs through sets of social relations that coalesce at particular moments to form ‘places’. Places are junctions or points of intersection within networks of social relations. I use the work of Bakhtin (1981; 1986) and Bourdieu (1977; 1991) to illustrate how, in each of the three places, language operates as a socio-ideological system that is divided, in flux and differentially empowered. This work on language as a social system was put into conversation with Lefebvre’s (1991) spatial theory, producing tools that were used as lenses through which to interpret the ethnographic fieldwork. What emerged was the centrality of the workings of language as a social system at Rosemary Gardens High School, Youth Amplified and amongst the Doodvenootskap. The control desired by educators, combined with the bureaucratic forces that restrict spontaneity in their teaching practices, resulted in the use of highly prescribed language forces dominating dialogic interactions at Rosemary Gardens High School. The different cultural influences and historical traditions, which produce the Doodvenootskap, led to the group reclaiming and reinventing varieties of language. At times this produced more sufficiently interactive forms of dialogic learning, amongst this group, and on other occasions they merely reiterated the words of others, without reflection or rigorous thought. Critical pedagogy, at Youth Amplified, laid the foundations for multiple contrasting perspectives and different linguistic forms to manifest. In the media and in the imaginary of the South African middle and upper classes, schools in neighbourhoods that were formerly reserved for ‘Black’ and working-class ‘Coloured’ children are generally perceived to be dysfunctional places. Young people who live in the neighbourhoods in which these schools are located, are assumed to learn very little. Research with youth from Rosemary Gardens discovered that this kind of negative portrayal is only one view of a multi-faceted set of stories. On a daily basis, young people from Rosemary Gardens use language in interactions with peers and adults, exchanges that shape their consciousness and influence how they make sense of the multiple social worlds which they partially produce.