Research

Several optimization problems in engineering do not have the necessary structure to be tackled by traditional optimization methods.  Indeed, the functions and the constraints defining the problem are often given in the form of a black-box (usually a computer code that reads some input and returns some output values).  The use of these black-boxes may be expensive, inaccurate and may fail for internal reasons to the box, thereby not allow derivative estimation. Professor Audet's research focusses on pattern search methods (GPS and MADS). A second aspect of his research efforts deal with exact global optimization methods for problem classes in which the structure is well defined and known, but nonconvex, such as bilevel, quadratic, bilinear and interger programming.  The structuring of these problems is identified and exploited in order to develop efficient algorithms for these problems. 

Professor Audet is interested in algorithmic development, convergence analysis and their applications.