Ronnie Fecso

second year master's student in ml

■ Computer Vision Intern at Vialytics, Stuttgart, Germany Apr 2022 – Aug 2022 Trained convolutional neural networks for the detection of deformations in cobblestone roads. Used pre-trained ResNet and Xception for object detection and image classification tasks and U-Net for semantic segmentation tasks. Implemented classical computer vision techniques like image augmentations as a pre-processing step and applied sound machine learning practices like using callback functions and running grid search. Also ran an experiment to test the effect of adding artificially deformed images as a way to increase training data size. In another project, applied Fourier transform, followed by auto-encoder network, followed by histogram-based gradient boosted regressor to predict unevenness of roads from accelerometer data. Model won data science hackathon. ■ Research Assistant at the Consumer Financial Protection Bureau, DC, USA Jun 2019 – Aug 2021 Providing quantitative research support in the Office of Research on the Financial Institutions and Regulatory Policy section. -Wrote Python script to automatically collect google search results of bank transaction data. Used this clearer textual data to create a neural network natural language processing model (in Python with Keras and nltk packages) that classified transactions into 4 categories with 86% accuracy. Saved countless hours by not requiring someone to manually google and classify 18000 transactions. -Wrote Postgresql queries to pull 100M row dataset that I then cleaned and aggregated. I used this data to plot loan denial rates over time for a research project examining bank discrimination. -Wrote queries in R using RODBC library to gather and aggregate bank and credit union financial activity data from internal PostgreSQL servers. Created R Shiny app which displayed an interactive US map of banks and credit unions and their assets, name, location, and international wire transfer counts when hovered over. Allowed user to control which institutions were plotted based on asset size and wire transfer count. -Ran time series analyses to study effect of major mortgage disclosure policy on volume of loan originations before and after rule implementation. Analyses and plots were published in Market Effects chapter of CFPB rule re-assessment. ■ Senior Co-Op With Anthem Inc., Blacksburg, VA, USA Aug 2018 – Dec 2018 Data scientist on a BI team trying to identify predictors/indicators of highly uninsured geographic areas. -Established a multi-step data cleaning and dimension reduction protocol on 1300 variable data set. -Applied Lasso and stepwise regression to identify 17 strongly predictive variables which fell into three categories: economic, demographic, and transportation. -Communicated the comparative predictive power of those 17 variables using random forest. -Reported results to VP-level employees in a 12 page technical report. ■ Teaching Assistant for Data Analytics and Viz., Blacksburg, VA, USA Aug 2018 – Dec 2018 Assisted students with their weekly homeworks (in Python using Jupyter Notebook) during office hours and through email. Walked students through web scraping, rigorous data cleaning and normalization, visualization using matplotlib, and unsupervised learning techniques like K-means clustering. Also met regularly with the instructor to give my input on what material to teach and how to teach it. Input homework grades weekly. Proctored final exam. ■ Data Science Fellow at Health and Human Services, Rockville, MD, USA Jun 2018 – Aug 2018 Accepted as one of 37 fellows out of 900 applicants. Placed at HHS as a data scientist. -Established novel data-driven health policy tools with the Chief Medical Officer and Office of IT to highlight agency’s efforts in combatting the opioid epidemic. Provided decision support to CMO. -Developed an interactive dashboard and mapping tool in Tableau. Dashboard had two maps side-by-side. Left map showed HHS opioid-related investments. Right map showed baseline opioid epidemic data like hospitalizations, death rate, and overdoses, etc. Communicated high dimensional, spatial-temporal relationships between policy efforts and opioid-health outcomes. -Presented dashboard at end of summer to HRSA (division of HHS) Administrator who said afterwards that he wants the project to continue. -In the following January, interviewed multiple candidates for the following 2019 Fellowship. ■ Honors Research - Data Science for the Public Good, Arlington, VA, USA May 2017 – Aug 2017 Helped develop evidence-based policy with grad students using R and SQL on open data at the Social Decision Analytics Laboratory -Presented actionable results to Alexandria City government addressing racial inequality -Using R, automated web scraping script which gathered info on 400,000 open source software projects to better understand economic impact of open source software (NSF funded project) -Perturbed sensitive, administrative data while still maintaining its usability

Université Jean Monnet, St. Etienne, France Aug 2021 – Jul 2023 ■ M.S. in Machine Learning and Data Mining ■ Erasmus semester at TU Wien - Fall 2022 Virginia Tech, Blacksburg, VA, USA Aug 2015 – May 2019 ■ B.S. in Computational Modeling and Data Analytics (CMDA) and Math Minor ■ Final GPA: 3.41