Speaker: Dr Reinier Diaz Millan
Federal Institute of Goias Title: On the splitting optimization problem with enlargement Date and time: 11:30am, Thursday 16 November 2017 Location: Room T121, Mt Helen Campus (this is at Federation University, RMIT crowd please connect via Visimeet).
In this paper, we present two approximate versions of the forward-backward splitting method for solving the minimization problem. In both cases, the objective function is the sum of two convex functions, maybe not differentiable. The algorithms involve, at each iteration, inexact evaluations of the backward operator and approximate subgradients of the functions (namely: the ε-subgradients). The first method considers an absolutely summable error criterion, whereas the second method uses a relative error criterion recently introduced for approximating proximal operators. Various stepsize rules are considered, including both diminishing and non-vanishing stepsizes, and convergence in objective values and convergence to a neighbourhood of the optimal set are obtained. The convergence analysis of the two methods shares underlying elements. Read more
Time and Location: Thursday, 19 October, 10:30 AM, T121, T Building, Mt Helen Campus (Federation University)
Title: Hyperbolic polynomials and semidefinite programming
Abstract: Hyperbolic polynomials are multivariate homogeneous polynomials with certain real-rootedness properties, such as the determinant restricted to symmetric matrices. First identified in the context of PDEs, they now play a role in numerous areas, including theoretical computer science, probability theory, and optimisation. Associated with any hyperbolic polynomial is a convex `hyperbolicity’ cone. The resulting conic optimisation problems are called hyperbolic programs.
Please contact Julien Ugon regarding the Visimeet connection. RMIT crowd please make your own arrangements with AGR booking/Visimeet connection.
Time and Location: Thursday, August 31 at 11:30 AM–12:30 PM, T121, T Building, Mt Helen Campus
Title: Decomposition and duality for Stochastic Integer Programs
Abstract: Stochastic Integer Programming is a powerful modelling tool for representing decision problems which incorporate elements of uncertainty. As compared to ordinary Integer Programs their scenario-based structure readily lends itself to decomposition and parallel computation techniques. Read more
Dr Deiderik Roijers from Vrjje Universiteit, Belgium is visiting Peter Vamplew at CIAO, Federation University. He will give two talks next week, both available via Visimeet (RMIT folk please make your own arrangements, AGR is not pre-booked for this).
14 August 2017 – Dr Deiderik Roijers, Vrjje Universiteit, Belgium Topic: Interactive Multi-objective Reinforcement Learning
1.30pm, Room T121, T Building, Mt Helen Campus or via the Visimeet System. Note different time slot. Download Poster
17 August 2017 – Dr Deiderik Roijers, Vrjje Universiteit, Belgium Topic: Policy/Value Reuse in Incremental Multi-objective planning and Learning
11.30am in Room T121, T Building, Mt Helen Campus or via the Visimeet system. Download poster