Patrick Emedom-Nnamdi

Patrick Emedom-Nnamdi, Timothy R. Smith, Jukka-Pekka Onnela et al.

We propose a nonparametric additive model for estimating interpretable value functions in reinforcement learning, with an application in optimizing postoperative recovery through personalized, adaptive recommendations. While reinforcement learning has achieved significant success in various domains, recent methods often rely on black-box approaches such as neural networks, which hinder the examination of individual feature contributions to a decision-making policy. Our novel method offers a flexible technique for estimating action-value functions without explicit parametric assumptions, overcoming the limitations of the linearity assumption of classical algorithms. By incorporating local kernel regression and basis expansion, we obtain a sparse, additive representation of the action-value function, enabling local approximation and retrieval of nonlinear, independent contributions of select state features and the interactions between joint feature pairs. We validate our approach through a simulation study and apply it to spine disease recovery, uncovering recommendations aligned with clinical knowledge. This method bridges the gap between flexible machine learning techniques and the interpretability required in healthcare applications, paving the way for more personalized interventions.

Patrick Emedom-Nnamdi, Abram L. Friesen, Bobak Shahriari et al.

Standard approaches to sequential decision-making exploit an agent's ability to continually interact with its environment and improve its control policy. However, due to safety, ethical, and practicality constraints, this type of trial-and-error experimentation is often infeasible in many real-world domains such as healthcare and robotics. Instead, control policies in these domains are typically trained offline from previously logged data or in a growing-batch manner. In this setting a fixed policy is deployed to the environment and used to gather an entire batch of new data before being aggregated with past batches and used to update the policy. This improvement cycle can then be repeated multiple times. While a limited number of such cycles is feasible in real-world domains, the quality and diversity of the resulting data are much lower than in the standard continually-interacting approach. However, data collection in these domains is often performed in conjunction with human experts, who are able to label or annotate the collected data. In this paper, we first explore the trade-offs present in this growing-batch setting, and then investigate how information provided by a teacher (i.e., demonstrations, expert actions, and gradient information) can be leveraged at training time to mitigate the sample complexity and coverage requirements for actor-critic methods. We validate our contributions on tasks from the DeepMind Control Suite.