OCKHAM

The OCKHAM team develops machine learning techniques and signal processing algorithms with the main objective of endowing them with solid theoretical foundations, physical interpretability and resource-efficiency.

With a culture rooted at the interface of signal processing and machine learning, the team’s expertise leverages the notion of parsimony and its structured variants – noticeably via graphs. Indeed, sparsity plays a fundamental role to warrant the identifiability of decompositions in latent spaces, such as inverse problems in high dimensional signal processing, and it also allows the development of distributed algorithms to learn from highly compressed data representations with privacy guarantees. Sparsity on graphs also gives rise to  techniques for semi-supervised learning in difficult settings. A major challenge is to leverage these ideas to ensure not only resource-efficient methods, but also explainable decisions and interpretable learnt parameters.