This dissertation develops and analyzes active learning algorithms for binary classification problems. In passive (non-active) learning, a learner uses a random sample of labeled examples from a fixed distribution to select a hypothesis with low error. In active learning, a learner receives only a sample of unlabeled data, but has the option to query the label of any of these data points. The hope is that the active learner needs to query the labels of just a few, carefully chosen points in order to produce a hypothesis with low error. The first part of this dissertation develops algorithms based on maintaining a version space---the set of hypotheses still in contention to be selected. The version space is specifically designed to tolerate arbitrary label noise and model mismatch in the agnostic learning model. The algorithms maintain the version space using a reduction to a special form of agnostic learning that allows for example-based constraints; this represents a computational improvement over previous methods. The generalization behavior of one of these algorithms is rigorously analyzed using a quantity called the disagreement coefficient. This algorithm is shown to have label complexity that improves over that of previous methods, and matches known label complexity lower bounds in certain cases. The second part of this dissertation develops algorithms based on simpler reductions to agnostic learning that more closely match the standard abstraction of supervised learning procedures. The generalization behavior of these algorithms are also analyzed in the agnostic learning model, and are shown to have label complexity similar to the version space methods. Therefore, these algorithms represent qualitative improvements over version space methods, as strict version space methods can be risky to deploy in practice. The first of these algorithms is based on a relaxation of a version space method, and the second is based on an importance weighting technique. The second algorithm is also shown to automatically adapt to various noise conditions that imply a tighter label complexity analysis. Experiments using this algorithm are also presented to illustrate some of the promise of the method

We present a simple, agnostic active learning algorithm that works for any hypothesis class of bounded VC dimension, and any data distribution. Our algorithm extends a scheme of Cohn, Atlas, and Ladner to the agnostic setting, by (1) reformulating it using a reduction to supervised learning and (2) showing how to apply generalization bounds even for the non-i.i.d. samples that result from selective sampling. We provide a general characterization of the label complexity of our algorithm. This quantity is never more than the usual PAC sample complexity of supervised learning, and is exponentially smaller for some hypothesis classes and distributions. We also demonstrate improvements experimentally.

Pre-2018 CSE ID: CS2007-0898

Unsupervised estimation of latent variable models is a fundamental problem central to numerous applications of machine learning and statistics. This work presents a principled approach for estimating broad classes of such models, including probabilistic topic models and latent linear Bayesian networks, using only second-order observed moments. The sufficient conditions for identifiability of these models are primarily based on weak expansion constraints on the topic-word matrix, for topic models, and on the directed acyclic graph, for Bayesian networks. Because no assumptions are made on the distribution among the latent variables, the approach can handle arbitrary correlations among the topics or latent factors. In addition, a tractable learning method via $\ell_1$ optimization is proposed and studied in numerical experiments.

Social media (SoMe) refers to a variety of virtual platforms used to enhance sharing of information. To evaluate the influence of SoMe with regards to views and downloads of published dermatology articles, we conducted a retrospective study from July 2020-March 2021 examining articles published on Instagram and Twitter under Dermatology Online Journal (DOJ) accounts and compared these with type-matched and issue-matched articles that were not posted on social media. During this time period, 163 total articles of the three types used for social media (Case Report, Case Presentation, and Photo Vignette) were published in DOJ and 15 were promoted via SoMe. Utilization of SoMe demonstrated a significant (P<0.0001) positive effect with regards to both views (175.5±16.4) and downloads (31.5±4.0) over matched articles not published on SoMe. Similar trends illustrating the positive effect of SoMe on readership have been previously observed in the field of dermatology as well as other medical specialties. Most direct accessions to articles arrived via Instagram rather than Twitter, diverging from previous studies on SoMe use in medical journals. Social media, in particular Instagram, can be a successful platform to enhance the exposure of peer-reviewed medical information.

This work considers a computationally and statistically efficient parameter estimation method for a wide class of latent variable models-including Gaussian mixture models, hidden Markov models, and latent Dirichlet allocation-which exploits a certain tensor structure in their low-order observable moments (typically, of second- and third-order). Specifically, parameter estimation is reduced to the problem of extracting a certain (orthogonal) decomposition of a symmetric tensor derived from the moments; this decomposition can be viewed as a natural generalization of the singular value decomposition for matrices. Although tensor decompositions are generally intractable to compute, the decomposition of these specially structured tensors can be efficiently obtained by a variety of approaches, including power iterations and maximization approaches (similar to the case of matrices). A detailed analysis of a robust tensor power method is provided, establishing an analogue of Wedin's perturbation theorem for the singular vectors of matrices. This implies a robust and computationally tractable estimation approach for several popular latent variable models.

The EGFR-mediated signaling pathways are important in a variety of cellular processes, including cell migration and wound re-epithelialization. Intracellular trafficking of EGFR is critical for maintaining EGFR surface expression. Galectin-3, a member of an animal lectin family, has been implicated in a number of physiological and pathological processes. Through studies of galectin-3-deficient mice and cells isolated from these mice, we demonstrated that the absence of galectin-3 impairs keratinocyte migration and skin wound re-epithelialization. We have linked this pro-migratory function to a crucial role of cytosolic galectin-3 in controlling intracellular trafficking and cell surface expression of EGFR after EGF stimulation. Without galectin-3, the surface levels of EGFR are markedly reduced, and the receptor accumulates diffusely in the cytoplasm. This is associated with reduced rates of both endocytosis and recycling of the receptor. We have provided evidence that this previously unreported function of galectin-3 may be mediated through interaction with its binding partner Alix, which is a protein component of the ESCRT (endosomal sorting complex required for transport) machinery. Our results suggest that galectin-3 is potentially a critical regulator of a number of important cellular responses through its intracellular control of trafficking of cell surface receptors.

The breakdown of triglycerides, or lipolysis, is a tightly controlled process that regulates fat mobilization in accord with an animal's energy needs. It is well established that lipolysis is stimulated by hormones that signal energy demand and is suppressed by the antilipolytic hormone insulin. However, much still remains to be learned about regulation of lipolysis by intracellular signaling pathways in adipocytes. Here we show that galectin-12, a member of a β-galactoside-binding lectin family preferentially expressed by adipocytes, functions as an intrinsic negative regulator of lipolysis. Galectin-12 is primarily localized on lipid droplets and regulates lipolytic protein kinase A signaling by acting upstream of phosphodiesterase activity to control cAMP levels. Ablation of galectin-12 in mice results in increased adipocyte mitochondrial respiration, reduced adiposity, and ameliorated insulin resistance/glucose intolerance. This study identifies unique properties of this intracellular galectin that is localized to an organelle and performs a critical function in lipid metabolism. These findings add to the significant functions exhibited by intracellular galectins, and have important therapeutic implications for human metabolic disorders.

BACKGROUND: Gamma sensory stimulation may reduce AD-specific pathology. Yet, the efficacy of alternating electrical current stimulation in animal models of AD is unknown, and prior research has not addressed intensity-dependent effects. METHODS: The intensity-dependent effect of gamma electrical stimulation (GES) with a sinusoidal alternating current at 40 Hz on Aβ clearance and microglia modulation were assessed in 5xFAD mouse hippocampus and cortex, as well as the behavioral performance of the animals with the Morris Water Maze. RESULTS: One hour of epidural GES delivered over a month significantly (1) reduced Aβ load in the AD brain, (2) increased microglia cell counts, decreased cell body size, increased length of cellular processes of the Iba1 + cells, and (3) improved behavioral performance (learning & memory). All these effects were most pronounced when a higher stimulation current was applied. CONCLUSION: The efficacy of GES on the reduction of AD pathology and the intensity-dependent feature provide guidance for the development of this promising therapeutic approach.

Background

Neurogenesis is significantly impaired in the brains of both human patients and experimental animal models of Alzheimer's disease (AD). Although deep brain stimulation promotes neurogenesis, it is an invasive technique that may damage neural circuitry along the path of the electrode. To circumvent this problem, we assessed whether intracranial electrical stimulation to the brain affects neurogenesis in a mouse model of Alzheimer's disease (5xFAD).

Methods and results

We used Ki67, Nestin, and doublecortin (DCX) as markers and determined that neurogenesis in both the subventricular zone (SVZ) and hippocampus were significantly reduced in the brains of 4-month-old 5xFAD mice. Guided by a finite element method (FEM) computer simulation to approximately estimate current and electric field in the mouse brain, electrodes were positioned on the skull that were likely to deliver stimulation to the SVZ and hippocampus. After a 4-week program of 40-Hz intracranial alternating current stimulation (iACS), neurogenesis indicated by expression of Ki67, Nestin, and DCX in both the SVZ and hippocampus were significantly increased compared to 5xFAD mice who received sham stimulation. The magnitude of neurogenesis was close to the wild-type (WT) age-matched unmanipulated controls.

Conclusion

Our results suggest that iACS is a promising, less invasive technique capable of effectively stimulating the SVZ and hippocampus regions in the mouse brain. Importantly, iACS can significantly boost neurogenesis in the brain and offers a potential treatment for AD.

Diffuse large B-cell lymphoma (DLBCL) is the most common non-Hodgkin lymphoma and an aggressive malignancy. Galectin-3 (gal-3), the only antiapoptotic member of the galectin family, is overexpressed in DLBCL. While gal-3 can localize to intracellular sites, gal-3 is secreted by DLBCL cells and binds back to the cell surface in a carbohydrate-dependent manner. The major counterreceptor for gal-3 on DLBCL cells was identified as the transmembrane tyrosine phosphatase CD45. Removal of cell-surface gal-3 from CD45 with the polyvalent glycan inhibitor GCS-100 rendered DLBCL cells susceptible to chemotherapeutic agents. Binding of gal-3 to CD45 modulated tyrosine phosphatase activity; removal of endogenous cell-surface gal-3 from CD45 with GCS-100 increased phosphatase activity, while addition of exogenous gal-3 reduced phosphatase activity. Moreover, the increased susceptibility of DLBCL cells to chemotherapeutic agents after removal of gal-3 by GCS-100 required CD45 phosphatase activity. Gal-3 binding to a subset of highly glycosylated CD45 glycoforms was regulated by the C2GnT-1 glycosyltransferase, indicating that specific glycosylation of CD45 is important for regulation of gal-3-mediated signaling. These data identify a novel role for cell-surface gal-3 and CD45 in DLBCL survival and suggest novel therapeutic targets to sensitize DLBCL cells to death.