Peng Liao Email and Phone Number

• Built an online reinforcement learning algorithm for use in mobile health that continuously learns and improves the treatment policy as the user experiences the interventions. • Collaborated with software engineers to implement the reinforcement learning algorithm in an experiment that aimed to help patients with hypertension improve physical activity.• Developed a sample-efficient reinforcement learning algorithm to identify the treatment policy that leads to optimal long-term performance and established a strong theoretical guarantee. • Designed a new robust criterion for policy optimization that can ensure uniform performance improvement over unknown planning horizons and initial distributions. • Worked as teaching assistant and guest lecturer for a graduate-level course, Sequential Decision Making (STAT 234). Advised 10 + class projects (algorithmic trading, social science, etc.).

• Developed a new approach for conducting statistical inference about the long-term performance of treatment policies using historical data collected under possibly different/unknown policies. • Designed a practical Thompson-Sampling contextual bandit algorithm that combines with a mixed-effect model to adaptatively pool data across users to speed up learning.

• Proposed new experimental designs for optimizing digital health interventions. Designed power and sample size calculation methods for detecting proximal treatment effects. • Designed a novel strategy that uses forecasting model to spread out interventions uniformly across high-risk times under a soft budget constraint. • Collaborated with researchers from computer science, engineering, and the behavioral sciences at Mobile Sensor Data-to-Knowledge (MD2K) to deploy the algorithm that determines the timing of prompts to perform stress-relaxation exercises in a smoking cessation study.