James Ferlez

James Ferlez Postdoctoral Scholar Resiliant Cyber-Physical Systems Laboratory (RCPSL) University of California, Irvine jferlez@uci.edu

Research

My current research activites are primarily focused on obtaining feedback controllers with correctness guarantees; this includes both correct-by-design methodolgies (e.g. AReN) and formal verification (e.g. Fast BATLLNN).

More broadly, I am interested in a number of adjacent and not-so-adjacent topics: neural networks; machine learning; formal methods (especially modal logic and aspects of process algebra); stochastic control; and reinforcement learning.

My research is summarized in this poster for the Meet the Faculty Candidates poster session at the 2022 Conference on Decision and Control (CDC).

Selected Publications

For a full list, please see my Google Scholar page.

Under Review

Assured Neural Network Architectures for Control and Identification of Nonlinear Systems. James Ferlez and Yasser Shoukry. 2021. PDF.

Recently Published

Polynomial-Time Reachability for LTI Systems with Two-Level Lattice Neural Network Controllers. James Ferlez and Yasser Shoukry. 2022. PDF.

Safe-by-Repair: A Convex Optimization Approach for Repairing Unsafe Two-Level Lattice Neural Network Controllers. Ulices Santa Cruz, James Ferlez and Yasser Shoukry. 61th Annual Conference on Decision and Control (CDC) 2022 (to appear). PDF.

Fast BATLLNN: Fast Box Analysis of Two-Level Lattice Neural Networks. James Ferlez, Haitham Khedr and Yasser Shoukry. ACM International Conference on Hybrid Systems: Computation and Control (HSCC), April 2022, pp 1 - 11. 0.1145/3501710.3519533.

Bounding the Complexity of Formally Verifying Neural Networks: A Geometric Approach. James Ferlez and Yasser Shoukry. 60th Annual Conference on Decision and Control (CDC) 2021. PDF.

PEREGRiNN: Penalized-Relaxation Greedy Neural Network Verifier. Haitham Khedr, James Ferlez and Yasser Shoukry. International Conference on Computer Aided Verification (CAV) 2021. SpringerLink; Pre-print PDF.

AReN: assured ReLU NN architecture for model predictive control of LTI systems. James Ferlez and Yasser Shoukry. Proceedings of the 23rd International Conference on Hybrid Systems: Computation and Control (HSCC) 2020. PDF.

Software

My personal GitHub page is here; the RCPSL GitHub page is here.

FastBATLLNN

Education

• Ph.D., Electrical Engineering, University of Maryland, College Park.

• M.S., Electrical Engineering, Pennsylvania State University, University Park, PA.

• B.S., Electrical Engineering, Pennsylvania State University, University Park, PA.