Our minds make inferences that appear to go far beyond standard data science. Whereas people can learn richer representations and use them for a wider range of learning tasks, machine learning algorithms have been mainly employed in a stand-alone context, constructing a single function from a table of training examples. In this talk, I shall touch upon a view on data science that can help capturing these human learning aspects by combining high-level languages and databases with statistical learning, optimisation, and deep learning. High-level features such as relations, quantifiers, functions, and procedures provide declarative clarity and succinct characterisations of the data science problem at hand. This helps reducing the cost of modelling and solving it. Putting probabilistic deep learning into the data science stack, it even paves the way towards one of my dreams, the automatic data scientist — an AI that makes data analysis and reporting accessible to a broader audience of non-data scientists.
This talk is based on joint works with many people such as Carsten Binnig, Martin Grohe, Zoubin Ghahramani, Martin Mladenov, Alejandro Molina, Sriraam Natarajan, Robert Peharz, Cristopher Re, Karl Stelzner, Martin Trapp, Isabel Valera, and Antonio Vergari.