Monday, April 24, 2017 - 14:05 , Location: Skiles 005 , Prof. George Mohler , IUPUI Computer Science , Organizer: Martin Short
In this talk we focus on classification problems where noisy sensor measurements collected over a time window must be classified into one or more categories. For example, mobile phone health and insurance apps take as input time series from the accelerometer, gyroscope and GPS radio of the phone and output predictions as to whether the user is still, walking, running, biking, driving etc. Standard approaches to this problem consist of first engineering features from statistics of the data (or a transform) over a window and then training a discriminative classifier. For two applications we show how these features can instead be learned in an end-to-end modeling framework with the advantages of increased accuracy and decreased modeling and training time. The first application is reconstructing unobserved neural connections from Calcium fluorescence time series and we introduce a novel convolutional neural network architecture with an inverse covariance layer to solve the problem. The second application is driving detection on mobile phones with applications to car telematics and insurance.
Series: CDSNS Colloquium
One dimensional discrete Schrödinger operators arise naturally in modeling the motion of quantum particles in a disordered medium. The medium is described by potentials which may naturally be generated by certain ergodic dynamics. We will begin with two classic models where the potentials are periodic sequences and i.i.d. random variables (Anderson Model). Then we will move on to quasi-periodic potentials, of which the randomness is between periodic and i.i.d models and the phenomena may become more subtle, e.g. a metal-insulator type of transition may occur. We will show how the dynamical object, the Lyapunov exponent, plays a key role in the spectral analysis of these types of operators.
Series: Combinatorics Seminar
Various parameters of many models of random rooted trees are fairly well understood if they relate to a near-root part of the tree or to global tree structure. In recent years there has been a growing interest in the analysis of the random tree fringe, that is, the part of the tree that is close to the leaves. Distance from the closest leaf can be viewed as the protection level of a vertex, or the seniority of a vertex within a network. In this talk we will review a few recent results of this kind for a number of tree varieties, as well as indicate the challenges one encounters when trying to generalize the existing results. One tree variety, that of decreasing binary trees, will be related to permutations, another one, phylogenetic trees, is frequent in applications in molecular biology.
Friday, April 21, 2017 - 15:00 , Location: Skiles 254 , Adrian P. Bustamante , Georgia Tech , Organizer:
A classical theorem of Arnold, Moser shows that in analytic families of maps close to a rotation we can find maps which are smoothly conjugate to rotations. This is one of the first examples of the KAM theory. We aim to present an efficient numerical algorithm, and its implementation, which approximate the conjugations given by the Theorem.
Series: ACO Student Seminar
Beginning with Szemerédi’s regularity lemma, the theory of graph decomposition and graph limits has greatly increased our understanding of large dense graphs and provided a framework for graph approximation. Unfortunately, much of this work does not meaningfully extend to non-dense graphs. We present preliminary work towards our goal of creating tools for approximating graphs of intermediate degree (average degree o(n) and not bounded). We give a new random graph model that produces a graph of desired size and density that approximates the number of small closed walks of a given sparse graph (i.e., small moments of its eigenspectrum). We show how our model can be applied to approximate the hypercube graph. This is joint work with Santosh Vempala.
Series: Professional Development Seminar
A conversation with Adam Fox, former GT postdoc who secured his "dream job" as a tenure-track assistant professor at Western New England University, but who recently moved into industry as a Data Scientist.
Series: Stochastics Seminar
We consider rooted subgraph extension counts, such as (a) the number of triangles containinga given vertex, or (b) the number of paths of length three connecting two given vertices. In 1989 Spencer gave sufficient conditions for the event that whp all roots of the binomial random graph G(n,p) have the same asymptotic number of extensions, i.e., (1 \pm \epsilon) times their expected number. Perhaps surprisingly, the question whether these conditions are necessary has remained open. In this talk we briefly discuss our qualitative solution of this problem for the `strictly balanced' case, and mention several intriguing questions that remain open (which lie at the intersection of probability theory + discrete mathematics, and are of concentration inequality type). Based on joint work in progress with Matas Sileikis
Series: Analysis Seminar
A well-known elementary linear algebra fact says that any linear independent set of vectors in a finite-dimensional vector space cannot have more elements than any spanning set. One way to obtain an analog of this result in the infinite dimensional setting is by replacing the comparison of cardinalities with a more suitable concept - which is the concept of densities. Basically one needs to compare the cardinalities locally everywhere and then take the appropriate limits. We provide a rigorous way to do this and obtain a universal density theorem that generalizes many classical density results. I will also discuss the connection between this result and the uncertainty principle in harmonic analysis.
Series: Research Horizons Seminar
In Fall 2017 I will teach `Random Discrete Structures', which is an advanced course in discrete probability and probabilistic combinatorics. The goal of this informal lecture is to give a brief outline of the topics we intend to cover in this course. Buzz-words include Algorithmic Local Locasz Lemma, Concentration Inequalities, Differential Equation Method, Interpolation method and Advanced Second Moment Method.
Series: Geometry Topology Seminar
Let S be a Riemann surface of type (p,1), p > 1. Let f be a point-pushing pseudo-Anosov map of S. Let t(f) denote the translation length of f on the curve complex for S. According to Masur-Minsky, t(f) has a uniform positive lower bound c_p that only depends on the genus p.Let F be the subgroup of the mapping class group of S consisting of point-pushing mapping classes. Denote by L(F) the infimum of t(f) for f in F pseudo-Anosov. We know that L(F) is it least c_p. In this talk we improve this result by establishing the inequalities .8 <= L(F) <= 1 for every genus p > 1.