Farzane Ezzati

Graduate Research Assitant, PhD Student

(2025) Riemannian ADMM for Binary Optimization


 I developed a Manifold-based modified ADMM algorithm tailored for large-scale or distributed binary-constrained optimization problems. The method is evaluated on challenging MIP-LIB NP-hard classified instances, binary distributed optimization instances, and large scale ML feature selection problems to assess convergence behavior and solution quality for various applications. 
  • Across the MIP test set, the approach achieved up to 90% alignment with global MIP solutions in 80% of cases, while delivering up to 50% reduction in computational time, demonstrating its potential as an efficient alternative for difficult binary constrained problems.
  •  Across the binary distributed optimization problems, the method consistently reached the optimal objective value for all test instances
  • The application to large-scale ML feature selection is currently under development, with implementation and evaluation in progress.