Deep learning: Unlocking the potential

The main goal of Project S seeks to promote and support the use of deep learning within the CRC. We will gather tools for analyzing deep neural networks—especially methods for comparing models with behavioral and neural data. We also seek to synthesize findings from across the CRC into a common theoretical framework based on deep learning. Specifically, we will test whether Prediction, Valuation and Categorization can be framed as different learning objectives. We will compare supervised, unsupervised and reward-based learning methods to develop unifying models of the “cardinal mechanisms” of perception.