Shanshan Lao
DARE Project: Observation of Optima Hopping in Simulated Annealing and Particle Swarm Optimization
Project Supervisor: Stephen Chen
The performance of many optimization algorithms can decrease as the dimensionality of the search space increases. I hope that this research can show the world a different way to understand this problem.
Project Description:
Computational Intelligence, soft computing, and nature-inspired algorithms such as Artificial Neural Networks have dominant results in many important real-world applications. However, a detailed understanding of how these algorithms work beyond their natural metaphors is often lacking. Detailed studies show that several nature-inspired algorithms function in code in ways that are quite different from the imagery created by their metaphors. Simulated Annealing is one of the first and most famous optimization algorithms, and its design mimics many features of the physical process of annealing. Recent studies show that the search trajectory does not follow chemical reaction paths, but that the technique instead degrades to ""optima hopping"". Similar behaviours have also been observed in Particle Swarm Optimization -- one of the most popular techniques in current use. The DARE student will have the opportunity to participate in field-altering research that breaks down the operation of these algorithms to show that the visualized search trajectories based on chemical reaction paths and the flight paths of birds do not actually occur. Instead, when applied to the domain of optimization, both algorithms degrade into optima hopping. These experiments will involve extensive coding (mostly Matlab), data collection, and data analysis.The Dean’s Award for Research Excellence (DARE) - Undergraduate enables our students to meaningfully engage in research projects supervised by LA&PS faculty members. Find out more about DARE.