Browsing by Subject "Active Learning"
Now showing 1 - 5 of 5
Results Per Page
Sort Options
Item Active and Adaptive Techniques for Learning over Large-Scale Graphs(2019-05) Bermperidis, DimitriosBehind every complex system be it physical, social, biological, or man-made, there is an intricate network encoding the interactions between its components. Learning over large-scale networks is a challenging field, and practical methods must combine scalability in computations to cope with millions of nodes associated possibly with large amounts of meta-information; along with sufficient versatility to capture the elaborate structure and dynamics of the complex phenomena under study. There is also a need for modeling expressiveness to ensure accurate learning, along with transparency and interpretability that will shed light on the overall system understanding, and will provide valuable insights about its function. Approaches to learning over networks must also defend against adversarial behavior, thus remaining operational even under severely adverse circumstances. The contribution of the present thesis lies at addressing the aforementioned challenges by developing simple yet versatile algorithmic solutions focused on core graph-learning tasks. To tackle active sampling for semi-supervised node classification, a novel framework is proposed in order to guide the sampling of informative nodes. The proposed framework is tailored to Gaussian-Markov random fields, and relies on the notion of maximum expected-change to select the most informative node to be labeled. Interestingly, several existing methods for active learning are subsumed by the proposed approach. Focusing on the node classification task, a generalized yet highly scalable diffusion-based classifier is developed, where each class diffusion is adaptive to the graph structure and the underlying label distribution. Adaptability is further leveraged for the node embedding task. As node embedding is naturally viewed as a low-rank factorization of a node-to-node similarity matrix, a versatile approach is introduced to learn the similarity matrix of a given graph with minimal computational overhead, and in a fully unsupervised manner. Extensive experimentation using both synthetic graphs as well as numerous real networks demonstrates the effectiveness, interpretability and scalability of the proposed methods. More importantly, the process of design and experimentation sheds light on the behavior of different methods and the peculiarities of real-world data, while at the same time generates new ideas and directions to be explored.Item Active Learning in STEM & Biology Learning and Teaching in the Laboratory Context(2021-07) Gonsar, NgawangThis three-paper dissertation addresses the experience and the implementation of evidence-based learning practices in science, technology, engineering, and mathematics (STEM)/biology education. Study 1 explored instructional strategies and student perceptions and preferences for various teaching practices in graduate and undergraduate classrooms across three STEM colleges. The study revealed that students desired more time for active learning practices and wanted fewer lectures than they currently experienced. Upon closer probing, findings suggest that educators should employ various active learning practices in their classrooms. Finally, the study provides suggestions for instructors teetering on the brink of adoption to leap into active learning.Study 2A and 2B narrowed the focus to learning in groups, which is the most utilized active learning strategy in biology courses. These studies examined how grouping strategies (self-selected vs instructor-assigned academically heterogeneous groups) impact first-year biology students' experience, performance, and cooperative learning participation in a biology laboratory course with extensive group work through a mixed-methods approach. There were similar effects on student perceptions from intervening in group strategies. However, students found substantial value in their group experiences in developing both academic and social skills. At the same time, students experienced diminishing concerns regarding their group members over time. When examining cooperative learning, there were many similarities but a greater frequency of cooperative learning elements when controlled for teacher's influence and the curriculum activity. There was also a small difference in the scores of assignments completed as a group. These findings in totality have implications on how instructors can best form groups that maximize student learning while improving students’ biology laboratory experience. The study findings suggest that once pedagogical approach and curriculum are controlled, there is evidence that academically heterogeneous groups, as opposed to self-selected groups, allow for more cooperative learning opportunities for first-year biology students.Item Data-driven Channel Learning for Next-generation Communication Systems(2019-10) Lee, DonghoonThe turn of the decade has trademarked the `global society' as an information society, where the creation, distribution, integration, and manipulation of information have significant political, economic, technological, academic, and cultural implications. Its main drivers are digital information and communication technologies, which have resulted in a "data deluge", as the number of smart and Internet-capable devices increases rapidly. Unfortunately, establishing information infrastructure to collect data becomes more challenging particularly as communication networks for those devices become larger, denser, and more heterogeneous to meet the quality-of-service (QoS) for the users. Furthermore, scarcity in spectral resources due to an increased demand for mobile devices urges the development of a new methodology for wireless communications possibly facing unprecedented constraints both on hardware and software. At the same time, recent advances in machine learning tools enable statistical inference with efficiency as well as scalability in par with the volume and dimensionality of the data. These considerations justify the pressing need for machine learning tools that are amenable to new hardware and software constraints, and can scale with the size of networks, to facilitate the advanced operation of next-generation communication systems. The present thesis is centered on analytical and algorithmic foundations enabling statistical inference of critical information under practical hardware/software constraints to design and operate wireless communication networks. The vision is to establish a unified and comprehensive framework based on state-of-the-art data-driven learning and Bayesian inference tools to learn the channel-state information that is accurate yet efficient and non-demanding in terms of resources. The central goal is to theoretically, algorithmically, and experimentally demonstrate how valuable insights from data-driven learning can lead to solutions that markedly advance the state-of-the-art performance on inference of channel-state information. To this end, the present thesis investigates two main research thrusts: i) channel-gain cartography leveraging low-rank and sparsity; and ii) Bayesian approaches to channel-gain cartography for spatially heterogeneous environment. The aforementioned research thrusts introduce novel algorithms that aim to tackle the issues of next-generation communication networks. Potential of the proposed algorithms is showcased by rigorous theoretical results and extensive numerical tests.Item Image classification with minimal supervision(2011-06) Joshi, Ajay JayantWith growing collections of images and video, it is imperative to have automated techniques for extracting information from visual data. A primary task that lies at the heart of information extraction is image classification, which refers to classifying images or parts of them as belonging to certain categories. Accurate and reliable image classification has diverse applications { web image and video search, content based image retrieval, medical image analysis, autonomous robotics, gesture-based human computer interfaces, etc. However, considering the large image variability and typically high-dimensional representations, training predictive models requires substantial amounts of annotated data, often provided through human supervision { supplying such data is expensive and tedious. This training bottleneck is the motivation for development of robust algorithms that can build powerful predictive models with little training or supervision. In this thesis, we propose new algorithms for learning with data, particularly focusing on active learning. Instead of passively accepting training data, the basic idea in active learning is to select the most informative data samples for the human to annotate. This can lead to extremely efficient allocation of resources, and results in predictive models that require far fewer training samples compared to the passive setting. We first propose an active sample selection criterion for training large multi-class classifiers with hundreds of categories. The criterion is easy to compute, and extends traditional two-class active learning to the multi-class setting. We then generalize the approach to handle only binary (yes / no) type feedback while still performing classification in the multi-class domain. The proposed modality provides substantial interactive simplicity, and makes it easy to distribute the training process across many users. Active learning has been studied from two different perspectives: selective sampling from a pool, and query synthesis; both perspectives o#11;er different tradeoffs. We propose a formulation that combines both approaches while leveraging their individual strengths resulting in a scalable and efficient multi-class active learning scheme.Experimental results show efficient training of classification systems with a pool of a few million images on a single computer. Active learning is intimately related to a large body of previous work on experiment design and optimal sensing { we discuss the similarities and key differences between the two. A new greedy batch-mode sample selection algorithm is proposed that shows substantial benefits over random batch selection, when iterative querying cannot be applied. We finally discuss two applications of active selection: i) active learning of compact hash codes for fast image search and classification, and ii) incremental learning of a classifier in a resource-constrained environment to handle changing scene conditions. Throughout the thesis, we focus on thorough experimental validation on a variety of image datasets to analyze strengths and weaknesses of the proposed methods.Item Machine Learning for Deep Brain Stimulation(2020-02) Grado, LoganDeep brain stimulation (DBS) is an effective treatment for a variety of neurological disorders, including Parkinson’s disease (PD). However, the success of DBS relies on selecting stimulation parameters which relieve symptoms while simultaneously avoiding stimulation-induced side-effects. Currently, DBS is programmed through a time-intensive trial-and-error process in which the clinician systematically evaluates stimulation settings, requiring hours of effort and multiple patient visits. Additionally, advances in DBS lead technology and stimulation algorithms are adding additional free parameters, further increasing the difficulty of programming these devices. This doctoral thesis advanced the programming of DBS arrays by: (1) developing the slid- ing windowed infinite Fourier transform (SWIFT), an efficient method of extracting oscillatory neural features which can be used to program DBS systems, (2) developing the Bayesian adaptive dual controller (ADC), a type of Active Learning DBS which can be used to learn optimal stimulation parameters, and (3) demonstrating the ef- ficacy of the Bayesian ADC in an animal model of PD. The primary findings of this dissertation suggest that the Bayesian ADC is capable of efficiently and autonomously learning stimulation parameters for DBS in order to optimize a selected biomarker. Furthermore, it was demonstrated that parameters learned by the Bayesian ADC performed as well as control parameters identified through a standard trial-and-error programming process. Together, these results suggest that the Bayesian ADC should be clinically translatable for tuning DBS in future studies.