shadow_tr

Colloquia

Fall 2018 Colloquia

Check back soon for more information on the computer science seminar series. Unless otherwise noted, the seminars meet on Mondays at 3pm in Stanley Thomas 302. If you would like to receive notices about upcoming seminars, you can subscribe to the announcement listserv.

Aug 28

Interdisciplinary Project Presentations

CS PhD Students TuLANE University

Please join us for interdisciplinary project presentations by the following Tulane computer science PhD students:

• Majid Mirzanezhad • Cody Licorish

This event will be held on Tuesday, 8/28/2018, from 3:30 - 4:30 p.m. in Stanley Thomas, Room 316. Please note the special weekday, time, and venue for this event.

Title of presentation by M.  Mirzanezhad: Fast Geometric Simplification of Maps and its Application in Geographic Information Systems

Abstract:
Geographic Information Systems (GIS) helps people visualize and create information from maps that can be used to make decisions and solve problems. In this talk, we particularly consider map simplification problem in ArcGIS that can make the view of a map nicer and avoid unnecessary computations on the map for any future needs. First, we go over some different variants of the problem and well-known algorithms for polyline simplification. We briefly provide some theoretical results and discuss their running time/space complexity and hardness. Next, we define our inspiration for the considered problem and we show how to exploit the existing algorithms for polyline simplification to obtain a simplification algorithm map. We also discuss our implemented toolbox in ArcGIS software with slightly faster query time which produces maps with respect to their scale with an appropriate quality of the view. Experimental results on the running time of our algorithm and the resulting maps obtained from two different datasets will be provided and compared in the end.

Title of presentation by C. Licorish: Comparison of Sampling Techniques for Photoacoustic Imaging

Abstract: Photoacoustic medical imaging can enhance traditional ultrasound imaging by adding diagnostic information. However, sampling hundreds of laser wavelengths for image reconstruction prohibits the practical use of photoacoustics from the clinic. This project compares different techniques for computing the best small set of wavelengths to use in photoacoustic applications. Both analytical and machine learning approaches are compared to the simple linear least-squares used in image reconstruction. Experiments are performed on images for measuring blood oxygen level.
Sept 17

Decision Making Over Combinatorially-Structured Domains

Andrea Martinez Hernandez Tulane University

Please join us for an interdisciplinary project presentation by Tulane computer science PhD student, Andrea Martin Hernandez.

Abstract: We consider a scenario where a user has to make a set of correlated decisions and we propose a computational model of the deliberation process. We assume the user compactly expresses her preferences as soft constraints. We design a sequential procedure which decomposes the complex decision task by using Decision Field Theory (DFT) to model deliberation on each variable. We compare our sequential approach to one in which a single deliberation is made over the set of complex objects. Our results show that the sequential approach outperforms the non-sequential one in terms of execution time and returns choices of similar quality. Finally, we consider three important effects which have been observed in human decision making: the similarity, attraction and compromise effect. These are modeled effectively by DFT in the case of non-structured alternatives. We show that our approach captures these phenomena for complex decisions which are decomposed in a combinatorial structure.
Oct 1

Establishing Trust Through Programming With Proofs

Brigitte Pientka McGill University

Abstract: Software systems should be robust, reliable, and predictable. A key step towards a safe and trustworthy infrastructure is verifying compliance of the software with a formal software policy. However, verifying complex safety properties in existing programming and proof environments can be costly and time consuming, as we need to repeatedly build up low-level infrastructure from scratch and using it in formal developments is tedious and error prone.

The Beluga project aims to change the way we develop and implement safe software systems by extending a general purpose programming and proof language with the ability to directly represent and manipulate formal systems and proofs. To specify formal systems and represent derivations within them, Beluga provides a sophisticated infrastructure based on the logical framework LF; to reason about formal systems, Beluga provides a dependently typed functional language for implementing inductive proofs about derivation trees as recursive functions following the Curry-Howard isomorphism. Key to this approach is the ability to model derivation trees that depend on a context of assumptions using a generalization of the logical framework LF, i.e. contextual LF which supports first-class contexts and simultaneous substitutions.

Our experience has demonstrated that Beluga enables direct and compact mechanizations of the meta-theory of formal systems, in particular programming languages and logics.

About the Speaker: Brigitte Pientka is an Associate Professor in the School of Computer Science at McGill University, Montreal, Canada. Her main research interests lie in the area of foundations of programming languages, type theories, and logic. Currently her research focus is on building a type-theoretic foundation for reasoning about formal systems.

B. Pientka received her Ph.D. from Carnegie Mellon University (Pittsburgh, USA). Recent honors include a Humboldt Research Fellowship (2016 - 2018) and an Accellerator Award (2012 - 2015) from the Natural Sciences and Engineering Research Council of Canada given to researchers who have an established, superior research program that is highly rated in terms of originality and innovation. She is serving on the editorial board of the Journal of Functional Programming and ACM Transactions on Computational Logic. She is also a member of the steering committee of the European Symposium on Programming (ESOP) and the Symposium on Principles of Programming Languages (POPL).
Oct 8

Machine Intelligence Projects at the Naval Research Laboratory-Stennis Space Center (NRL-SSC)

Paul Elmore Naval Research Lab

Abstract: Machine intelligence (MI) has been extensively applied recently in geospatial applications. At NRL we have projects that involve MI in spatial bathymetry. One project uses machine learning (ML) to classify seafloor features such as underwater seamounts and ridges. Specifically, variations of decision trees known as random forests and extremely randomized trees are used for classification. Training and testing is done for bathymetry data from active seafloor areas of the Eastern Pacific. I will discuss these experiments and significant findings. Another current NRL project uses ML to enhance the resolution/accuracy of seafloor data obtained by satellite altimetry gravity measurements. Here we employ twelve super resolution methods trained from combinations of gravity and multi-beam sonar data. The techniques include, among others, the use of neural net variants and a random forest technique, super-resolution forests. Last, a new project to begin in FY-18 will investigate the application of deep learning to image classification. However, without understanding how deep learning arrives at a solution there is no guarantee that these networks will transition from a controlled laboratory environment to a fieldable system. This particular project will develop the incorporation of explanatory rule based methodologies into neural networks by embedding fuzzy inference systems into deep learning networks.

About the Speaker: Paul A. Elmore, Ph.D. is a research physicist for the Marine Geosciences Division of the Naval Research Laboratory (NRL), Stennis Space Center, MS. He has 2 patents and over 45 technical articles in print. He is an IEEE Senior Member. Since 2007, he has been a principle investigator of NRL research projects on bathymetry data fusion, interpolation, uncertainty estimation and databasing. Recent work includes research on Dempster-Shafer, fuzzy-set theories for geospatial information uncertainty, and game theory. He has served as Chair of the General Bathymetry Charts of the Oceans (GEBCO) technical presentation sessions, GEBCO Science Day, and as convener at American Geophysical Union (AGU) meetings. Dr. Elmore received a B.S. (magna cum laude) in physics from Millsaps College in Jackson, MS in 1990 and a Ph.D. in physics from the University of Mississippi in 1996. His graduate work was in nonlinear acoustics, and he was a graduate fellow at the National Center for Physical Acoustics. He received a fellowship from the American Society of Engineering Education to pursue postdoctoral research at NRL from 1996-1998. He worked for the Naval Oceanographic Office from 1998-2001, returning to NRL in 2001. Dr. Elmore lives in the New Orleans area with his wife and children. His publications are online at: https://www.researchgate.net/profile/Paul_Elmore.
Oct 15

Assessing Transplant Compatibility using Statistical Learning

Loren Gragert Tulane Cancer Center

Abstract: The genes involved in the adaptive immune system are more diverse than any others in the human genome, enabling us to mount a response to an enormous diversity of pathogens. HLA molecules on the cell surface present antigens to T cells so that they can see inside the cell to detect infections or mutations. As pathogens evolve to evade detection, HLA genetic diversity is maintained at a high level, with over 18,000 alleles identified in human populations. Because the immune system uses HLA to distinguish "self" versus “nonself", HLA matching between donor and recipient improves outcomes of transplantation. Globally, bone marrow registries have recruited and HLA genotyped over 33 million volunteers to donate stem cells if they match a patient. Patients seeking a transplant are also tested for the presence of antibodies to HLA specificities. Because of technological limitations in these HLA assays, transplant matching algorithms must use statistical inference to select the best possible donor based on ambiguous or incomplete information. Recent advancements in transplant matching have required collaboration among the disciplines of immunology, population genetics, epidemiology, statistics, and computer science.

About the Speaker: Dr. Loren Gragert is an Assistant Professor in the Department of Pathology and Laboratory Medicine at Tulane University School of Medicine. He received his PhD in Biomedical Informatics and Computational Biology from the University of Minnesota in 2014. He was a bioinformatics scientist at National Marrow Donor Program / Be The Match for 12 years before joining Tulane in 2015. His research focuses on adapting immunogenetics datasets and informatics tools developed initially for bone marrow registries for new applications in the kidney allocation system.
Oct 22

Topic

Speaker Institution

Abstract: TBA
Oct 29

Topic

Speaker Institution

Abstract: TBA
Nov 5

PhD Prospectus Presentation

Jaelle Scheuerman Tulane University

Abstract: TBA
Nov 12

Topic

Speaker Institution

Abstract: TBA
Nov 19

Topic

Speaker Institution

Abstract: TBA
Nov 26

Topic

Speaker Institution

Abstract: TBA
Dec 3

Topic

Speaker Institution

Abstract: TBA

303 Stanley Thomas Hall, New Orleans, LA 70118 504-865-5785 compsci@tulane.edu