Speaker: Devin Willmott

Title: Generative Neural Networks in Semi-Supervised Learning

Abstract: Semi-supervised learning is a relatively new machine learning concept that seeks to use both labeled and unlabeled data to perform supervised learning tasks. We will look at two network types with some promising applications to semi-supervised learning: ladder networks and adversarial networks. For each, we will discuss the motivations behind their architectures & training methods, and derive some favorable theoretical properties about their capabilities.

Title:   Matrix Factorization Techniques for Recommender Systems
Abstract: Recommendation Systems apply Information Retrieval techniques to select the online information relevant to a given user. Collaborative Filtering (CF) is currently most widely used approach to build Recommendation System. To address this issue, the collaborative filtering recommendation algorithm is based on singular value decomposition (SVD) . How the SVD works to make recommendations is presented in this master talk.

Jonathan Proctor will be giving a Master's Talk.  He will be presenting the paper

Learning About When and Where from Imagery
Speaker: Nathan Jacobs, University of Kentucky
Abstract:

Every day billions of images are uploaded to the Internet. Together they provide many high-resolution pictures of the world, from panoramic views of natural landscapes to detailed views of what someone had for dinner. Many are tagged with when and where the picture was taken, thus providing an opportunity to better understand how the appearance of objects and scenes varies with respect to location and time. This talk describes my work in using learning-based methods to extract geo-spatial properties from imagery. In particular, I will focus on two recent research thrusts: using deep convolutional neural networks to geo-calibrate social network imagery and using such imagery to build geo-temporal models of human appearance.

BIO:

Nathan Jacobs earned a PhD in Computer Science at Washington University in St. Louis (2010). Since then, he has been an Assistant Professor of Computer Science at the University of Kentucky. Dr. Jacobs' research area is computer vision; his specialty is developing learning-based algorithms and systems for processing large-scale image collections. His is a recipient of an NSF CAREER award, and his research has been funded by ARMY-SMDC, ARL, DARPA, Google, IARPA, NGA, and NIH. His current focus is on developing techniques for mining information about people and the natural world from geotagged imagery, including images from social networks, publicly available outdoor webcams, and satellites.

Speaker: Luis Sordo Vieira
Title: The benefits of elliptic curve cryptography
Abstract: We will introduce the basis of elliptic curve cryptography.  Roughly speaking ECC is based on the group structure of the points defined on an elliptic curve over a finite field and the difficulty of solving the discrete log problem. The applications are many, such as signature verification and pseudo random generators. No knowledge of algebraic geometry is required.

Computing Exponentials of Essentially Non-negative Matrices with Entry-wise Accuracy
Speaker: Qiang Ye, University of Kentucky
Abstract:

A real square matrix is said to be essentially non-negative if all of its off-diagonal entries are non-negative. In this talk, I will present new perturbation results and algorithms that demonstrate that the exponential of an essentially non-negative matrix can be computed with entrywise relative accuracy.

Learning Algorithms for Restricted Boltzmann Machines
Speaker: Devin Willmott, University of Kentucky
Abstract: Restricted Boltzmann machines (RBMs) have played a central role in the development of deep learning. In this talk, we will introduce the theoretical framework behind stochastic binary RBMs, give motivation and a derivation for the most commonly used RBM learning algorithm (contrastive divergence), and prove some analytic results related to its convergence properties.

Abstract: As biology has become a data-rich science, more biological phenomena have become amenable to modeling and analysis using mathematical and statistical methods. At the same time, more mathematical areas have developed applications in the biosciences, in particular algebra, discrete mathematics, topology, and geometry. This talk will present some case studies from algebra and discrete mathematics applied to the construction and analysis of dynamic models of biological networks. Some emerging themes will be highlighted, outlining a broader research agenda at the interface of biology and algebra and discrete mathematics. No special knowledge in any of these fields is required to follow the presentation.

Title: "A Matrix Analysis of Centrality Measures"

Abstract:  When analyzing a network, one of the most basic concerns is identifying the "important" nodes in the network. What defines "important" can vary from network to network, depending on what one is trying to analyze about the network. In this paper by Benzi and Klymko several different centrality measures, methods of computing node importance, are introduced and compared. We will see that some centrality measures give more information about the network on a local scale, while others help to analyze on a more global scale. In particular, the paper analyzes the behavior of these measures as we let the parameters defining them approach certain limits that appear to be problematic.

Title: Optimality of the Neighbor Joining Algorithm and Faces of the Balanced Minimum Evolution Polytope

Abstract:  Balanced minimum evolution (BME) is a statistically consistent distance-based method to reconstruct a phylogenetic tree from an alignment of molecular data. In 2008, Eickmeyer, Huggins, Pachter, and myself developed a notion of the BME polytope, the convex hull of the BME vectors obtained from Pauplin's formula applied to all binary trees. We also showed that the BME can be formulated as a linear programming problem over the BME polytope.  The BME is related to the Neighbor Joining (NJ) algorithm, now known to be a greedy optimization of the BME principle. Further, the NJ and BME algorithms have been studied previously to understand when the NJ algorithm returns a BME tree for small numbers of taxa. In this talk we aim to elucidate the structure of the BME polytope and strengthen knowledge of the connection between the BME method and NJ algorithm. We first show that any subtree-prune-regraft move from a binary tree to another binary tree corresponds to an edge of the BME polytope. Moreover, we describe an entire family of faces parametrized by disjoint clades. We show that these clade-faces are smaller-dimensional BME polytopes themselves. Finally, we show that for any order of joining nodes to form a tree, there exists an associated distance matrix (i.e., dissimilarity map) for which the NJ algorithm returns the BME tree. More strongly, we show that the BME cone and every NJ cone associated to a tree T have an intersection of positive measure.  We end this talk with the current and future projects on phylogenomics with biologists in University of Kentucky and Eastern Kentucky University.  This work is supported by NIH.