This talk addresses three key topics within Dynamic Multi-Objective Optimization (DMO). Firstly, addressing the historical inconsistent and ill-informed usage of parameters that control the frequency and severity of dynamics; the subsequent introduction of a comprehensive experimental procedure for assessing their impacts within commonly used benchmark problems; and establishing meaningful and reproducible baseline performance using well-known (non-dynamic) evolutionary algorithms and basic dynamic responses, for the more informed and logical experimentation in DMO. Secondly, addressing a gap in available realistic DMO problem benchmarks in the combinatorial domain, we adapt the complex and interconnected Travelling Thief Problem, to include three different types of contextually justified dynamics. This effectively provides us with a bridge between simplistic benchmarks and the end-goal real-world application problem. A varying impact of initialization on obtained final solutions has been constructed in a Responsive Seeding Algorithm to demonstrate the possible utility of exploiting problem knowledge as a dynamic change response. Performance and efficiency are contrasted with adapted methods from the static TTP. Finally, DMO as a field is ever-expanding in the number of novel algorithms and applications and benchmarks. However, one overlooked problem class is multi-dynamic instances. These are problems which contain more than one dynamic component (within the objective functions or elsewhere) with independent frequencies and/or severity of change. An introduction to navigating the possible instance space, approaching experimentation, and visualizing measurements for this novel problem class is provided.
Bio: Daniel Herring is a final-year OPTIMA PhD student and Priestley Scholar working between the University of Melbourne and the University of Birmingham in the UK. Before this, he completed an MSci in Natural Sciences at the University of Exeter, UK, where his project involved using evolutionary algorithms to estimate neural pathway model parameters from EEG data in Alzheimer’s patients.
