Postdoc position in reinforcement learning and operations research
marti...@gmail.com
PROJECT MOTIVATION & GOALS: This project aims to harness the power of deep reinforcement learning (DRL) to solve online stochastic optimization problems (OSOPs). DRL has seen increasing success in solving tasks that involve sequential decision-making under uncertainty. Perhaps the most well-known application of DRL in this domain is gaming, where it has been used to train agents capable of superhuman performance in (among others) Chess, Go, Doom, Texas Hold'em Poker, StarCraft II, and many Atari classics. In the latter example, a single-agent architecture was able to outperform humans in nearly 49 different games. This feat is notable as these games differ in appearance, goals, rewards, actions, etc. Given the diversity of games on which this architecture has proved successful, one wonders whether it would perform comparably well on any game with a similar format. Should this premise hold, it presents the opportunity to use this (or a similar) agent architecture to solve OSOPs, provided that the OSOPs can be properly formatted as video games. We refer to this process as "gamification".
The main objective of this project is to show that gamification presents an opportunity to extend the achievements of the DRL community to solve hard OSOPs by using new problem formulations that apply a visual representation to classical problem models. This approach was successfully used to solve the vehicle routing problem with stochastic service requests (VRPSSR). However, more extensive research on more complex and realistic problems is needed to conclude on the prospect of gamification.
CONTEXT: The fellow will be based at the Inter-University Research Center on Enterprise Networks, Logistics, and Transportation (CIRRELT) in the city of Montreal (Canada). The fellow will work under the direction of Professors Martin Cousineau (HEC Montréal), Jorge E. Mendoza (HEC Montréal) and Amir-massoud Farahmand (University of Toronto). The fellow will receive a salary of 60k CAD/year (taxable) for at least one year. The position is renewable for up to a total of 3 years depending on results and fit. Additional funding may become available via excellence scholarships (e.g., from the schools, the research center, and the provincial or federal governments).
DESIRED QUALIFICATIONS: The ideal applicant should possess strong computer programming skills and demonstrate a deep understanding of deep reinforcement learning (DRL) techniques. Familiarity with operations research (OR) modeling tools and solution methods, including mathematical programming, decomposition techniques (such as column generation and Benders), and Markovian decision processes, is also beneficial. Additionally, experience in programming for video games would be highly advantageous. Effective communication in English is essential. Candidates holding a Ph.D. degree in operations research, management science, computer science, industrial engineering, or applied mathematics are encouraged to apply.
CONTACT: To apply, interested candidates should email Professors M. Cousineau (martin.c...@hec.ca) and J.E. Mendoza (jorge....@hec.ca) with the following attachments: an up-to-date CV; a copy of the Ph.D. dissertation and relevant publications; the names and contact information of two references. Please use “[Gamification] Postdoctoral application” as the subject of the email. The position will remain open until filled.