IRMER - Research initiative on recommendation methods
A corporate patronage for recommendation methods
The Research initiative on recommendation methods (IRMER - Initiative de Recherche autour des MEthodes de Recommendation) is a program of the PGMO funded by Criteo.
The focus of this IRMER initiative is recommendation methods, both from a theoretical and a practical viewpoints (developing new methods and evaluating their performance on real data).
Mathematicians and computer scientists from both the academic and industrial worlds can benefit from it. Projects are open to all academic researchers with no restrictions due to administrative or geographic location. They should be relevant to the field of data science (including machine learning, statistics, and computer science in relation to data analytics) and should be focused on providing new results concerning recommendation systems.
Suggested research agenda
- Axis 1: Low Dimensional Representations, Embeddings and Feature Selection
The first axis of this call for projects focuses on the development of any technique that can allow an efficient learning based on data of very high dimension, with an application to recommendation systems. We may mention deep learning, feature selections, etc., but these are just examples of techniques that could be useful. Any other approach is welcome.
- Axis 2: Efficient and Fast Optimization and Decision Making
This second axis concerns fast optimization techniques and algorithms than can be used to handle the high-frequency constraints of a real-world recommendation system.
- Axis 3: Repeated, Real Time Decision Making
The third axis focuses on real-life, practical repeated decision-making and how the actual constraints violate the usual assumptions and prevents the use of classical algorithms.
- Axis 4: Long-Term Interactions, Time Dependencies
This fourth line of research relies on taking into account time dependencies to devise fast, efficient algorithms for the optimization of the system and/or repeated decision making.