University of Ottawa

Published purpose

Al-powered design automation inevitably involves iterations between generating a candidate design and evaluating its quality. As a result, rapid evaluation of the candidate designs is critically needed. In practice, however, evaluating the quality of a design is usually carried out by a physical process that is resource intensive and/or time consuming. As such, this resource and time consumption can be a major bottleneck for Al-powered design automations. To overcome this limitation, a surrogate evaluation function or a "surrogate model", is often used to replace the evaluation process and a learning approach is taken to estimate the parameters of the model. More specifically, the surrogate model is learned from a set of designs for which quality measurements have been obtained from the original evaluation process. This Project focuses on studying the effectiveness and robustness of such a learning approach and developing novel methods for this purpose.