Patrick Roos, PhD
Ph.D. Computer Science, UMD
M.S. Computer Science, UMD
B.S. Data Science, College of Charleston
Railroad Technology and Information Services
Next Generation Estimated Time of Arrival System
Built a highly-accurate system to estimate when cargo and intermodal trains will arrive on the North American major rail lines. The approach is implemented with a graph-style database in HBase and machine learning models developed on Spark. Set up a secured EMR cluster for the analysis.
Large Multinational Construction Company
Deep Learning for Complex Construction Sequencing
Extracted features from highly complex datasets and developed a reinforcement learning approach to determine the sequence and install order for millions of items (pipe, steel, concrete) on a construction site in three-dimensional space. Used GPU clusters to optimize construction planning prior to construction start.
- Python data stack: scikit-learn, pandas, numpy, Jupyter, others
- Hadoop: Spark, MapReduce, Hive, HBase,
- Machine Learning: Supervised and Unsupervised Learning, Reinforcement learning, MCTS, Survival Analysis, CLV, Customer Segmentation, Customer 360, Recommendation Engines
- Amazon Web Services: EC2, ECS, RedShift, EMR, Lambda, S3
- Biomedical Informatics and Genomics: Bowtie, BLAST, Kraken