A Gene Expression Programming Approach to Designing CNN Architectures

As part of our course project for EECS 592: Foundations of AI at the University of Michigan, my peers and I tackled the problem of intelligently and automatically designing CNN architectures. We explore two evolutionary paradigms, namely (i) Cartesian Genetic Programming (which acts as a baseline), and (ii) Gene Expression Programming (GEP). For GEP, we also experiment with the inclusion/exclusion of dropout, as well as two different evolutionary algorithms: crossover mutation and the \((1+\lambda)\) evolutionary algorithm. More details about our work can be found in our poster and our report.