- Homepage
- About the Joint Centre
- Research Areas & Projects
- Researchers
- News & Events
- Open Positions
In this project we cover three research areas:
1) Large-scale convex optimization
Large-scale convex optimization for big data: Many classification problems that practitioners are facing have large dimensions: large size of observations p, large number of observations n, large number of classes k. Supervised learning algorithms based on convex optimization (such as the support vector machines or logistic regression) cannot yet achieve robustly the ideal complexity O(nk+np), i.e., the size of the data. To achieve this complexity, we propose several open avenues of research (whose impact on solving large problems with higher accuracy is immediate).
Specifically we intend to focus on i) Large-scale learning with latent structure, ii) Optimal computational-statistical trade-offs, iii) Large-scale learning with large number of classes : scaling in log(k), iv) Large-scale active learning, v) Large-scale learning with large number of classes : dealing with imbalance and vi) Robust stochastic approximation algorithms.
2) Large-scale combinatorial optimization
Many problems in computer vision or natural language processing involve large-scale combinatorial problems, for example involving graph cuts or submodular functions. We propose several open avenues of research at the interface between combinatorial and convex optimization (whose main impact is to enlarge the set of problems that can be solved through machine learning).
In this context we intend to work on i) Large-scale learning with large number of classes : leveraging the latent geometry; ii) Relationships between convex and combinatorial optimization; and iii) Structured prediction without pain.
3) Sequential decision making for structured data
Multi-Armed Bandit (MAB) problems constitute a generic benchmark model for learning to make sequential decisions under uncertainty. They capture the trade-off between exploring decisions to learn the statistical properties of the corresponding rewards, and exploiting decisions that have generated the highest rewards so far. In this project, we aim at investigating bandit problems with a large set of available decisions, with structured rewards. The project addresses bandit problems with known and unknown structure, and targets specific applications in online advertising, recommendation and ranking systems.
2020
Communication dans un congrès
2019
Article dans une revue
Communication dans un congrès
2017
Communication dans un congrès
Pré-publication, Document de travail
2016
Communication dans un congrès
Pré-publication, Document de travail
2015
Communication dans un congrès
2014
Article dans une revue
Communication dans un congrès
2013
Article dans une revue
2012
Communication dans un congrès
Rapport
2011
Communication dans un congrès
2006
Article dans une revue
