Nonparametric Policy Improvement in Continuous Action Spaces via Expert Demonstrations

Authors: Agustin Castellano, Sohrab Rezaei, Jared Markowitz, Enrique Mallada
Conference: Reinforcement Learning Conference

This paper introduces a novel nonparametric policy for deterministic MDPs that admits a policy improvement theorem. It constructs a value lower bound using expert demonstrations, enabling robust evaluation and a simple improvement mechanism. The proposed policy optimization algorithm guarantees monotonic improvement and high-probability performance.

🔗 View on OpenReview

Comparative Analysis of Data-Driven Predictive Control Strategies

Authors: Sohrab Rezaei, Ali Khaki-Sedigh
Conference: 2023 ICCIA | Publisher: IEEE
Date: December 20, 2023 | Pages: 1–6

This paper compares three key data-driven predictive control methods—DeePC, WKPC, and MFAPC—highlighting their theoretical bases, assumptions, and performance through a benchmark example. It offers insights into the trade-offs in applying these methods.

🔗 View on Google Scholar 🔗 View on ArXiv

Mini Jet Turbocharged Engine for Use in a Thermotronic Laboratory

Authors: Kamyar Nikzadfar, Mahdi Garivani, Amir H Zabihi, Sohrab Rezaei
Journal: Journal of Technology in Aerospace Engineering, Vol. 3, Issue 1, Pages 1–11
Publisher: ARI | Date: May 22, 2019

This project focuses on developing a turbojet engine prototype for laboratory use in control and measurement systems. It includes design, assembly, and performance analysis of the engine using the Brayton cycle, establishing infrastructure for aerospace control research.

🔗 View on Google Scholar