Research

We propose a new approach to the multiclass many-server queueing scheduling problem in the fluid limit, inspired by optimal transport theory. The reformulation yields the optimal scheduling policy within a general class of discipline scheduling policies over a finite horizon.

We are developing a deep reinforcement learning approach for integer-valued action spaces that constructs each action sequentially, rather than emitting it in a single one-shot decoder output.

A noncooperative dynamical model of elections is developed. We analyze the case of two candidates competing for a fixed, stationary electorate over a finite horizon. Numerical examples illustrate how voter distribution, campaign duration, and initial positions shape equilibrium candidate trajectories. We extend the model to incorporate credibility losses from policy repositioning.

Refines the classical Shapley–Folkman–Starr lemma to obtain a tighter estimate of the Hausdorff distance between a Minkowski sum and its convex hull, by treating all components jointly rather than separately. Applies the result to the course-allocation problem, a combinatorial assignment setting.