Computational Foundry Seminar Series

 

DATE & TIME: Tuesday 21st November 2017 at 2pm

LOCATION: CoFo Seminar Room, 909 Talbot Building

TITLE: Person Re-identification in Wide-Area Surveillance

AUTHORS: Professor Shishir Shaw

ABSTRACT: Person re-identification is the problem of identifying a person from an image, given a set of gallery images of different persons across different cameras or varying viewing angles. It has various potential applications in the areas of automated video surveillance and human computer interaction. However, the task of person re-identification poses considerable difficulties due to variations in illumination, viewpoint, pose and even occlusion. A typical system designed for person re-identification takes a probe image and a set of gallery images, and tries to find the closest match of the probe image amongst the gallery images. Usually, matches are obtained by finding the gallery image which has the least distance from or the greatest similarity score with the probe image.  In this talk, I present an overview of the problem in the context of wide-area surveillance and its challenges.  In addition, I will discuss different perspectives for addressing this problem.

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

DATE & TIME: Tuesday 14th November 2017 at 2pm

LOCATION: CoFo Seminar Room, 909 Talbot Building

TITLE: Computational Acoustic Levitation for Display Technologies

AUTHORS: Asier Marzo, Univeristy of Bristol

ABSTRACTProgrammed acoustic waves can levitate particles of a wide range of materials and sizes through air, water or biological tissues. New applications for Human-Computer Interaction (HCI) can be derived from the possibility of moving objects in mid-air to specific locations and even through obstacles. In the most basic configuration, we move particles over a surface to paint on sand or liquids at a distance and without contact. A more advanced, system positions a couple of objects in 3D allowing us to represent functions and positions of objects such as planes or asteroids. The ultimate goal is a display made of hundreds of levitating particles that move independently to form different shapes using computational acoustic waves

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

DATE & TIME: Tuesday 31st October 2017 at 2pm

LOCATION: CoFo Seminar Room, 909 Talbot Building

TITLE: Finding the Perfect Triangle

AUTHORS: Sean Walton

ABSTRACT: Simulation has changed the way engineers approach their work.  No longer forced to build physical models and perform real experiments, simulation allows them to test thousands of ideas to find the perfect design.

As simulations have become increasingly accurate, researchers have focused efforts on reducing how long they take to calculate.  In this talk I will explain how trying to reduce this time led me to a somewhat under used simulation technique and a 6-year search for the perfect triangle.

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

DATE & TIME: Tuesday 24th October 2017 at 2pm

LOCATION: CoFo Seminar Room, 909 Talbot Building

TITLE: Noncomputability in analysis

AUTHORS: Arno Pauly 

ABSTRACT: Many theorems in analysis state the existence of a certain object depending on some parameter. Each such theorem has an associated computational task: Compute the object from the parameter. From the viewpoint of a constructivist, these tasks are intricately linked to the meaningful truth of the theorems. From a pragmatic perspective, the applicability of a theorem to fields like physics or economics is tied to the solvability of the associated computational task.

We will see that many central theorems from analysis are actually not computable. Ironically, our prime example will be Brouwer's Fixed Point theorem. We can classify the extent of non-computability using the framework of Weihrauch degrees. Knowing the Weihrauch degree of a theorem entails a lot of information, for example whether the associated task might at least be probabilistically solvable, or whether it could be made solvable by adding a discrete advice parameter.

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

DATE & TIME: Tuesday 17th October 2017 at 2pm

LOCATION: CoFo Seminar Room, 909 Talbot Building

TITLE: Safety-complete Test Suites

AUTHORS: Wen-ling Huang and Jan Peleska (presentation by Jan Peleska) 

ABSTRACT: This presentation is about property-oriented testing. A novel safety-related variant of complete test suites for finite state machines is introduced. Under certain hypotheses which are similar to the ones used in the well-known complete testing methods like W-Method, Wp-Method, HSI-Method, or H-Method, the new method guarantees to uncover every safety violation, while erroneous outputs without safety-relevance may remain undetected. In well-defined situations that can be precisely pre-determined from the reference model, this leads to a substantial reduction of test cases in comparison to the size of the analogous W, WP, HSI, H-test suites. We advocate this new test strategy for situations, where exhaustive testing of the complete system is too expensive. In these cases, strong guarantees with respect to fault coverage should only be given for the errors representing safety violations, while it is considered as acceptable if less critical errors remain undetected. An original version of this material has been published at the ICTSS 2017 conference; in this talk, we present a refined test suite based on the H-method which can be shown to always produce less or equally many test cases as when applying the original H-method. We sketch how this strategy can be extended to safety-complete equivalence class testing for systems with infinite input domains but finitely many internal states and finite output domains.