Institute for Systems Research and Department of Electrical and Computer Engineering
Ant-based algorithms are proposed for the binary bridge selection problem, and a number of issues concerning their convergence properties discussed. For the simplest algorithm, with bridges of equal length, we identify the range of parameter values which yield reinforcement learning—All ants go on the same bridge!
The presentation will emphasize connections with urn models and the theory of stochastic approximations. In the second half of the talk we touch on implementation issues (e.g., finite memory) and the case of bridges with unequal lengths.