Babak Aslani - Healthcare Analytics

Scalable Optimization Frameworks for Large-Scale Decision Analytics in Healthcare

Modern healthcare decision problems, such as treatment planning, personalized medicine, and resource allocation, often involve high-dimensional optimization models with millions of decision variables and complex clinical constraints. Solving such problems efficiently and accurately is critical for supporting data-driven and evidence-based medical decisions. This research develops mathematically rigorous and computationally scalable frameworks for large-scale optimization in healthcare analytics. The proposed methodology integrates first-order and second-order optimization paradigms to enable warm-started interior-point methods that achieve faster convergence and higher accuracy compared to existing solvers. These hybrid methods combine the efficiency of gradient-based approaches with the robustness of second-order algorithms, providing both scalability and precision for complex clinical applications. From a managerial and practical standpoint, this work advances the field of predictive and prescriptive healthcare analytics, offering a foundation for improved decision support in areas such as personalized treatment design, hospital resource management, and operational planning. The accompanying open-source software, warmip, implements these ideas in an open-access Python package for warm-started interior-point quadratic programming, bridging theoretical innovation and practical deployment in healthcare optimization.

Funding Agency: Department of Medical Physics at MSKCC.

Role: Postdoctoral Research Fellow

Highlights

Warm-IP: A novel hybrid optimization framework combining Path-Following ADMM with a primal-dual Interior-Point Method (IPM).
Addresses key IPM limitations — improves warm-start capability and early-iteration efficiency.
Path-Following ADMM efficiently generates warm-starts along the central path, enabling fast IPM convergence.
Demonstrated superior performance on real-world and synthetic quadratic programming (QP) benchmarks.
Achieved 11× speedup and 53% less computation time compared to MOSEK, the fastest commercial solver.
Scales to large problems with up to 2 million variables and constraints.
Offers a scalable and accurate framework for optimization in healthcare, control, and operations research.

Publications

• Aslani, B., Tefagh, M., Jhanwar, G., Zarehpisheh, M., (2025). Warm-IP: A Path-Following ADMM Warm-start for Interior-Point Methods. (Submitted to INFORMS Journal on Computing).