Open Positions and Thesis Topics
Most decisions involve trade-offs. Individuals, organizations, and automated systems find decisions hard because objectives conflict, consequences unfold over time, and the environment is uncertain. Multi-objective optimization and reinforcement learning provide the formal tools to address these problems.
The topics below are open at all levels: BSc, MSc, PhD, and postdoctoral. All connect to active research lines in the group. If you are interested, write to carlos.hernandez at iimas.unam.mx with a brief description of your background and the direction that interests you.
Algorithmic Topics
Multi-objective reinforcement learning. Learning agents that optimize several conflicting objectives simultaneously and produce a set of policies that represents the different trade-off solutions.
Efficient approximation of the set of policies. Algorithms that cover the full set of trade-off policies at a computational cost close to that of training a single policy.
Optimization and decision making under uncertainty. Algorithms and statistical criteria that find and compare solutions when the environment is noisy, measurements are imperfect, or parameters vary.
Tracking trade-offs in changing environments. Methods that follow the set of trade-off solutions as the environment or the objectives change over time.
Interactive and preference-based optimization. Methods that incorporate human preferences into the search process and let decision makers guide and refine solutions progressively.
Constrained multi-objective optimization. Methods for problems in which some objectives act as thresholds that a solution must meet.
Archivers and set representations for multi-objective algorithms. Design and analysis of strategies that maintain representative solution sets during the search, with formal convergence properties.
Structure of policy spaces. Analysis of how the geometry of neural network policies shapes the search for trade-off solutions.
Benchmarks and evaluation protocols. Test problems and protocols that measure how algorithms handle noise, limited budgets, changing environments, and problem size.
Automatic algorithm configuration and design. Methods that select and configure algorithmic components for optimization problems, including the use of reinforcement learning for this task.
Applied Topics
Adaptive decision systems in finance. Multi-objective frameworks for credit, portfolio, and risk decisions that balance profitability, fairness, and transparency.
Multi-objective optimization for urban mobility and sustainability. Algorithms that support participatory planning processes and integrate community knowledge and nature-based solutions.
Logistics and operations under uncertainty. Multi-objective methods for routing, scheduling, and resource allocation problems with stochastic elements.
Robotics and control. Multi-objective learning of control policies for systems that operate in changing conditions, such as legged robots.