Dimensions, bits, and wows in accelerating materials discovery

In this book chapter, we discuss how the problem of accelerated materials discovery is related to other computational problems in artificial intelligence, such as computational creativity, concept learning, and invention, as well as to machine-aided discovery in other scientific domains. These connections lead, mathematically, to the emergence of three classes of algorithms that are inspired largely by the approximation-theoretic and machine learning problem of dimensionality reduction, by the information-theoretic problem of data compression, and by the psychology and mass communication problem of holding human attention. The possible utility of functionals including dimension, information [measured in bits], and Bayesian surprise [measured in wows], emerge as part of this description, in addition to measurement of quality in the domain.