Dr Michael Gutmann

Risk-averse design under uncertainty

Mathematical models are typically used to accelerate system design and optimization ahead of implementation. However, models are inherently inaccurate which can result in sub-optimal or non-functional performance when deployed. To mitigate this issue, we propose to combine Bayesian inference, Thompson sampling, and risk management to find optimal designs. Our approach uses data from unsuccessful designs to estimate the distribution of the model parameters and then employs risk-averse optimization to select design parameters that are expected to perform well given parameter uncertainty and system noise. We illustrate the approach by designing genetic circuits under uncertainty.