(Non-)Parametric Recoverability of Preferences and Choice Prediction

Simple functional forms for utility require restrictive structural assumptions that are often contrary to observed behavior. Even so, they are widely used in applied economic research. I address this issue using a two-part adaptive experimental design to compare the predictions of a popular parametric model of decision making under risk to those of non-parametric bounds on indifference curves. Interpreting the latter as an approximate upper bound, I find the parametric model sacrifices very little in terms of predictive success. This suggests that, despite their restrictiveness, simple functional forms may nevertheless be useful representations of preferences over risky alternatives.

Adaptive Experimental Design for Optimizing Combinatorial Structures

Scientists and engineers in diverse domains need to perform expensive experiments to optimize combinatorial spaces, where each candidate input is a discrete structure (e.g., sequence, tree, graph) or a hybrid structure (mixture of discrete and continuous design variables). For example, in hardware design optimization over locations of processing cores and communication links for data transfer, design evaluation involves performing a computationally-expensive simulation. These experiments are often performed in a heuristic manner by humans and without any formal reasoning. In this paper, we first describe the key challenges in solving these problems in the framework of Bayesian optimization (BO) and our progress over the last five years in addressing these challenges. We also discuss exciting sustainability applications in domains such as electronic design automation, nanoporous materials science, biological sequence design, and electric transportation systems.

Envelope Simulation: A New Adaptive Method for Dose Optimization

We present a flexible and general adaptive experimental design for dose finding clinical trials. This method can be applied in sterotypical settings such as determining a maximum tolerated dose, when the dose response relationship is complex, or when the response is an arbitrary quantitative measure. Our design generalizes dose finding methods such as the continual reassessment method (CRM), modified CRM, and estimation with overdose control (EWOC). Similar to those, our design requires a working mathematical model, which assures efficiency, low bias, and a higher fraction of dose selected near the optimum compared to purely operational designs. Unlike typical model based designs that always employ the same model, our method allows individual dose response models tailored to the circumstance. Simulations are also integral to our design allowing it to account for both known and unknown effects of dose on outcome. This method is applicable to general dose finding problems such as those encountered with modern targeted anti-cancer agents, immunotherapies, and titrations against biomarker outcome measures. The method can also support improvements in the design of the experiment being conducted by providing a platform for concurrent simulations to assess the influence of projected design points. Although we present the design in the context of clinical trials, it is equally applicable to experiments with non-human subjects.