Author(s): Steven Weijs
Linked Author(s): Steven Weijs
Keywords: No Keywords
Abstract: Building predictive models is akin to data compression: by describing the patterns found in observations in the most compact way, we can extrapolate to predict unseen observations. Minimizing description length of data is an underlying objective in both physics based and data driven approaches. An emerging perspective from Algorithmic Information Theory is presented, which could contribute to a consistent framework to provide guidance on the strengths and weaknesses of both approaches, and their combination.
Year: 2022