The recent development of recording techniques in neuroscience has led to an avalanche of high-dimensional neural data. For example, 2-photon imaging allows to routinely record from large populations of neurons simultaneously, and microelectrodes grant access to hundreds or thousands of neurons with millisecond resolution. A key principle for analyzing such data is to reduce their dimensionality onto i) dimensions that explain most of the variance in the data (“neural manifolds”), ii) dimensions that are most related to experimental variables such as the stimulus or the response of the animals (encoding/decoding), or iii) dimensions that best explain activity in other parts of the brain (communication subspaces).

The major goal of this project is to tailor corresponding state-of-the-art machine learning and data analysis techniques to the demands of the whole consortium, and to develop new methods where needed.