Roozbeh Yousefzadeh
PhD
Computer Science Department
University of Maryland
2108 Brendan Iribe Center
8125 Paint Branch Drive
College Park, MD 20742
U.S.A.
Email: my first name @cs.umd.edu

Tweet to @1optimizer

I will join Smita Krishnaswamy's Lab at Yale University, as a Postdoc.

I completed my PhD in the Department of Computer Science at University of Maryland, College Park. My research advisor and mentor at UMD was Professor Dianne P. O'Leary. My main research interests are scientific computing, machine learning, and solving hard optimization problems.

For my dissertation, I have been working on interpretation of neural networks and application of homotopy methods. We developed mathematical methods to study trained neural networks as nonlinear functions, interpret their behavior, study their decision boundaries, and to improve and debug them. Our research has implications for adversarial robustness of deep learning models and their generalization error, too. We also developed methods to refine the structure of trained networks using matrix conditioning.

Last summer, I was at the Los Alamos National Laboratory for their Applied Machine Learning Fellowship. I finished a project about interpretation of Gaussian graphical models for unsupervised learning of data gathered by the Mars rover. It is published in the proceedings of a SIAM conference.

I like to solve challenging optimization problems in different contexts. I am comfortable with nonlinear optimization problems, integer programming and linear programming. During my coursework, I took Scientific Computing I and II, Nonlinear Optimization II, Integer Programming, Sparsity and Machine Learning, Machine Learning, High Performance Computing, and several other courses.

I have worked on real-time optimization problems in Humanitarian Aid delivery systems and Freight transportation systems. We developed a probabilistic framework to optimize the decisions. Our formulation is nonlinear and non-convex, and we designed a homotopy algorithm to solve it in real-time.

In Fall 2018, I received the Graduate School's Outstanding Teaching Assistant Award. That semester, I was a TA for the course CMSC460 "Computational Methods".

I am also a licensed structural engineer in Maryland. Before studying computer science, I was a structural design engineer.

Here is a list of papers that I have worked on:

  1. Interpreting Neural Networks Using Flip Points
  2. Debugging Trained Machine Learning Models Using Flip Points
  3. Learning Diverse Gaussian Graphical Models and Interpreting Edges
  4. Investigating Decision Boundaries of Trained Neural Networks
  5. Refining the Structure of Neural Networks Using Matrix Conditioning
  6. Optimizing Real-time Decisions in Hierarchical Humanitarian Aid Delivery Systems
  7. A Probabilistic Framework and a Homotopy Method for Real-time Hierarchical Freight Dispatch Decisions

Roozbeh Yousefzadeh

Web Accessibility