Thursday, February 7, 2019, 2PM
138 Sawyer Conference Room
Abstract: Automated and semi-automated systems that derive actionable information from massive, heterogeneous datasets are essential for many applications, and the reasoning of such systems must be as clear as possible in order to earn our trust. Lattice data analytics, a current research project, is intended to extend the bounds of lattice theory and its applications. If successful, the information-awareness algorithms will lead to automated explanatory methods for machine learning. The algorithms, based on three novel lattice-theoretic concepts (the target/event lattice, the temporal poset of irreducibles, and the lattice entropy), are being designed to be capable of detecting structure in temporal, multivariate datasets. They will use the concept of a Dedekind-MacNeille completion to clarify predictive relationships between lattice nodes, which will represent real or virtual targets and/or events. It is expected that the algorithms will tolerate missing, messy, or otherwise incomplete data. The poset of irreducibles will be used to compress the data and to further enable the modified Dedekind-MacNeille completion algorithm to run on massive datasets in near-real time. A free, open-source Python toolkit will be made available to support the use and visualization of the lattice-theoretic data exploration and analytic algorithms, thereby enabling researchers and developers to rapidly produce systems that leverage the novel data analysis technique. The Carver2 target rating program, which includes a small dataset of partially ordered potential targets, is used to illustrate the functionality of the toolkit. Future work includes implementation of interactive lattice visualization tools and analysis of the algorithms using real datasets. It is hoped that collaboration with the Climate Change Institute will lead to the application of the lattice-theoretic data analysis algorithms to climate change datasets as well as to the development of interactive information visualizations of interest to both researchers and educators. Preliminary talks have laid the groundwork for collaboration with researchers in the Climate Change Institute, TIEMS (The International Emergency Management Society), and the HERACLES Project, an EU-funded program to protect cultural heritage sites from the effects of climate change.