Work to be presented at SPIE Defense and Commercial Sensing in April

Work related to our US Space Force’s UCRO project will be presented during the Algorithms, Technologies, and Applications for Multispectral and Hyperspectral Imaging meeting at SPIE DCS on April 23. The talk, Using neural networks to classify hyperspectral signatures of unresolved resident space objects, will discuss our work on using deep learning methods. The paper includes undergraduate Luis Cedillo and masters student Kevin Acosta as authors in addition to Drs. DeBlasio and Velez-Reyes who are the co-investigators on the project.

Luis Cedillo presents poster at UCRO 1.0

Undergraduate student Luis Cedillo will present work related to our project “Innovative Analysis of Spectra-Temporal Signatures using Machine Learning for Ground-Based Remote Sensing of Unresolved Resident Space Objects”, collaborative work with Kevin Acosta and Dr. Miguel Velez-Reyes (Electrical and Computer Engineering) at the University Consortium Research Opportunity meeting in Boulder, CO. This stands as our one year anniversary of the project funded by the US Space Force through the University Consortium Research Opportunity.

Research Focus

Dr. DeBlasio has recently moved to Carnegie Mellon University‘s Computational Biology Department as an Assistant Teaching Professor. As a result, his lab is currently in transition and not highly active. For details about Dr. DeBlasio’s previous work see his personal website:

Our group studies how to improve science by automating and optimizing the tools used by domain scientists. We do this primarily by making input specific parameter value choices which help to reduce false information introduced by using less than ideal (or default) parameter choice. Using a framework called Parameter Advising we are able to, without an increase in wall clock time in most cases, find parameter vectors that are much better then the defaults. This framework has been applied to both protein multiple sequence alignment and reference-based transcript assembly, but is very general and can be applied to domains both within and outside of computational biology.

Beyond the algorithm configuration problem, Dr. DeBlasio also has interests in hashing and sketching, primarily focused on minimizer schemes (also called winnowing schemes). Minimizer schemes are a method to represent long strings by some representative k-mer (k length substring) in order to improve the resource consumption of sequence analysis applications (such as genomic read mapping, or document similarity).