Physics Department Seminar
Dr. Julio J. Valdes
Research Scientist
National Research Council, Data Science for Complex Systems Group
Tuesday, January 23, 2018

Computational Intelligence and Physics: A Hand Shake

The purpose of the talk is to present an overview of Computational Intelligence approaches (a branch of Artificial Intelligence), as tools within  experimental and theoretical research in Physics. Computational Intelligence and Machine Learning techniques cover a broad domain of different areas, among them, neural networks, evolutionary computation, fuzzy logics, rough sets, probabilistic reasoning, kernel methods and others. Several topics related to the analysis of data and the formulation and study of first principles models in Physics will be discussed from a computational intelligence and machine learning perspective in the context of the Information Explosion and the Big Data scenarios. Modern developments in sensor, communication and computer technologies have revolutionized data acquisition by increasing the amount of information obtained from complex systems, and are received increasing attention. A related (overlooked) consequence has been the increasing degree of heterogeneity of the information obtained. Heterogeneous data refers to objects described by features of different nature (e.g. mixtures of numeric, qualitative (nominal), ordinal, interval, images, documents, signals, graphs, etc.). In addition to the complexity introduced by the heterogeneity of the attributes, the information usually is incomplete (missing values) and comes with different types and degrees of uncertainty. A heterogeneous dataset may contain hundreds, thousands or even millions of such objects. The discussion will cover i) working with heterogeneous data (data exploration, knowledge discovery, advanced visualization techniques), and ii) modeling (development of surrogate models, learning equations from data, enhancing first-principles models with data-driven models and the creation of hybrid models). Real world examples are presented for important operations in data analytics like classification, regression and data visualization using virtual, mixed and augmented reality techniques. An important objective is to stimulate a discussion about how to incorporate computational intelligence techniques within computational physics for both experimental and theoretical research.