NSIDC Cryosphere Seminar
Visual Analytics and Interactive Machine Learning for Geospatial Sciences by Dr. Morteza Karimzadeh, Assistant Professor, Department of Geography, CU Boulder
Machine learning is increasingly used in various stages of scientific inquiry, from data cleaning and fusion, to analysis and insight generation. The full realization of machine learning in many scenarios is still limited by the sparsity of labeled training data, which is expensive and difficult to generate. Even when available, labeled training datasets capture a snapshot in time and space, resulting in models that may not perform well under different conditions. Additionally, models may reflect the biases inherent in the training data. In this talk, I will present on multiple interactive visual analytics frameworks for the simultaneous labeling, learning and analysis of data in two different domains, namely streaming social media document analytics and feature selection in hyperspectral imagery for precision agriculture. Both represent cases with spatiotemporal heterogeneity and limited training data for building performant models. In presenting these visual analytics approaches, I will break down the underlying computational components and the interactive interfaces, and draw connections on how such approaches can be adopted in cryospheric data and research, as well as other domains utilizing multi-source, dynamic and streaming data.