We are always seeking for motivated students for pursuing a PhD in our team.
If you are interested, please send us your application, including a CV and a motivation letter per email to randopt-hiring AT inria.fr.
Right now (September 2023), we have a concrete offer for a PhD position in the context of our project “Learning to optimize”. Instead of developing an algorithm the tedious way “by hand” as it was done in successful (blackbox) optimization algorithms such as the CMA-ES, we see two principal ways to exploit machine learning techniques to come up with novel, potentially improved optimization algorithms. In the bottom up approach, we take a state-of-the-art algorithm and identify one or several update equations based on (i) its possible weaknesses or potentials of improvement and (ii) the possible interactions with the remaining updates from changing the equation. Then, we either learn a transformer for this identified equation or we learn an equation as its replacement from scratch. In the top down approach, we learn from scratch the entire algorithm, for example as a generative neural network for a time-changing sample distribution. In addition to the unconstrained non-convex, non-differentiable problem domain, we also plan to further explore the idea of learning to optimize for problems with equally practically relevant properties such as constrained, mixed-integer, and multiobjective optimization.
Expected starting date: early 2024
Applications welcome anytime (CV and motivation letter per email to randopt-hiring AT inria.fr).
