MCMC for High Energy X-Ray Radiography
DateThursday, November 1, 2018 - 12:30pm
AbstractImage deblurring via deconvolution can be formulated as a hierarchical Bayesian inverse problem, and numerically solved by Markov Chain Monte Carlo (MCMC) methods. Numerical solution is difficult because • Inconsistent assumptions about the data outside of the field of view of the image lead to artifacts near the boundary; and • The Bayesian inverse problem is high-dimensional for high-resolution images. The numerical MCMC framework I present addresses these issues. Boundary artifacts are reduced by reconstructing the image outside the field of view. Numerical difficulties that arise from high-dimensions are mitigated by exploiting sparse problem structure in the prior precision matrix.
Discrete Exterior Calculus in Solving Fractional Differential Equations
DateThursday, November 8, 2018 - 12:30pm
AbstractIn modeling natural occurrences, we often concede that we are close enough to reality. However in certain cases, integer-ordered differential equations actually fail to correctly simulate what we expect to physically occur. Some instances of this include groundwater flow, computer simulations of how cloth moves about underwater, and even the electric current in hearts affected by ischaemia. In this talk, we will discuss three main points—these motivating examples, fundamentals of fractional differential equations (FDE’s) in space, and Discrete Exterior Calculus (DEC), the method which we would like to apply to solving these equations.
Algorithmic approaches to identification and antibiotic susceptibility testing of pathogenic microbes
DateThursday, November 15, 2018 - 12:30pm
AbstractIn this talk, I'll first provide some context regarding the challenges of hospital-acquired infections, antibiotic resistance, and sepsis, a disease estimated to kill millions worldwide, and one of the leading causes of death in hospitalized patients. I'll then review traditional laboratory techniques for identifying pathogenic microbes and performing antibiotic susceptibility testing (AST), and how antibiotic susceptibility is reported in terms of the Minimum Inhibitory Concentration (MIC). For the remainder of the talk, I'll describe the technologies and algorithms used by the Accelerate Pheno system for rapid identification and AST. Identification involves automated fluorescent in-situ hybridization, or FISH, coupled with algorithmic analysis of microscopic images. For AST, the Pheno system extracts a number of features from time-lapse images of growing bacterial colonies exposed to antibiotics (e.g. morphology, division rates, and growth curves), which become the inputs to a suite of machine learning algorithms for determining MIC values.