Mauricio is a Computer Science PhD Candidate (recently passed his proposal defense) currently enrolled at the University of Delaware.
His areas of interest involve deep learning/machine learning and high performance computing, focusing on developing machine learning models applied to healthcare and accelerating computational science applications using high-level parallel programming models.
Mauricio is a member of the CRPL group at the University of Delaware and is advised by Dr. Sunita Chandrasekaran.
Project in collaboration with the Nemours Children's Hospital under Dr. Erin L. Crowgey (now at Incyte) and Prof. Adam Marsh, leveraging machine learning approaches to predict outcomes for rare diseases using transcriptomics, variant data, and electronic health records.
Worked as a solutions architect intern for NVIDIA on a project accelerating the annual energy production (AEP) calculation of a python-based wind farm simulation tool using Cupy and GPUs.
Conducted research on a machine learning project focused on predicting phenotype group on a Parkinson's Disease cohort using genomic variant data and methylation scores.
Oversaw responsibility for integrating the weather simulation NICAM-DC source code into the SPEChpc 2021 benchmark harness.
Worked as a summer intern for the National Cancer Institute (NCI) under Dr. Eric Stahlberg and Dr. George Zaki various Pilot 1 deep learning models of the ECP-CANDLE project showing scalability of multi-GPU training using Horovod and finding uncertainty quantification using dropout at inference as a Bayesian approximation.
In collaboration with Department of Chemistry & Biology’s Prof. Juan Perilla and under Prof. Sunita Chandrasekaran’s advisement, accelerated a chemical shift prediction algorithm that provides insight into protein secondary structure using the high-level parallel programming model OpenACC.
Journal publication about this project can be found here: https://doi.org/ 10.1371/journal.pcbi.1007877
In the era of accelerators, transitioning large scale applications to heterogenous systems that take advantage of the host-device execution model is the path to improve and optimize runtime performance looking ahead on the future of exascale computing.
The world of machine learning is thriving due to the expanding growth of big data. More information can be gathered, and more accurate models can be trained over large amounts of data. However, adding more data means the process becomes quite expensive. For this reason, computer scientists must begin looking at HPC techniques to efficiently reduce maximize performance while reducing the cost of these large data models.
• Mauricio H. Ferrato, Adam G. Marsh, Karl Franke, Benjamin J. Huang, E. Anders Kolb, Deborah DeRyckere, Douglas K. Graham, Sunita Chandrasekaran, and Erin L. Crowgey. ”Machine learning classifier approaches for predicting response to RTK-Type-III Inhibitors demonstrates high accuracy using transcriptomic signatures and ex vivo data.” Bioinformatics Advances,2023 https://doi.org/10.1093/bioadv/vbad034
• Eric Wright, Mauricio H. Ferrato, Alex J Bryer, Searles Robert, Juan R Perilla, Sunita Chandrasekaran, (2020) “Accelerating prediction of chemical shift of protein structures on GPUs: Using OpenACC.” PLOS Computational Biology 16(5): e1007877. https://doi.org/10.1371/journal.pcbi.1007877