Today, the world faces an unprecedented increase in the volume and speed of available data streams. Many applications must move from offline to sequential methods that can acquire, adapt to, and process data on the fly. At the same time, the data is becoming increasingly sophisticated. Traditional statistical assumptions such as stationarity (or i.i.d. data) are no longer satisfied. Designing efficient algorithms that can learn from data as it comes in with as few assumptions as possible is a significant challenge in today's machine learning. Harnessing the potential of these real-time data streams is the goal of online (streaming, recursive) learning. This project aims at building online boosting methods that learn from the data step by step, improving when observing more information. In other words, the goal is to adapt successful Boosting algorithms to the sequential paradigm. The project mixes theoretical research with applications on real data through interdisciplinary applications. It plans to design algorithms with robust theoretical guarantees and good practical performance.
PhD student: Paul LIAUTAUD
PhD supervisors: Olivier Wintenberger, Pierre Gaillard
Research laboratory: LPSM - Laboratoire de Probabilités Statistique & Modélisation