Ryan Socha Email & Phone Number

Greater Fayetteville, Ar Area

• Funded by the Air Force Research Lab to develop generative AI image-to-image models improving the quality of radar images of target military vehicles.• First authored “Assessing the value of phase to deep SAR ATR via non-inferiority testing” whichproposes a novel feature importance method to test the importance of noisy features via controlled trials on modified “placebo” datasets (currently under publication review).• Coauthored “Graph pretraining approach to utilize synthetic data for SAR ATR”, presented atSPIE Defense and Commercial Sensing’s April 2024 Conference.

Fayetteville, Arkansas, United States

• Learned the mathematics of traditional methods for computational fluid dynamics (CFD) modeling such as Direct Numerical Simulation, Large-eddy Simulation, and Reynolds-averaged Navier-Stokes models.• Conducted an extensive literature review on the strengths and weaknesses of machine learning models vis-à-vis traditional CFD models, with a particular focus on the advantages of Li et al. 2020's Fourier Neural Operator, and wrote a 10-page expository paper on the topic.• Authored a grant proposal to replicate and extend Li's results that was accepted and funded by the University of Arkansas' Honors College.• Also investigated theoretical extensions to the Generalized Negative Correlation Learning strategy of Buschjäger, Pfahler, and Morik 2020.

Fayetteville, Arkansas, United States

• Designed an overhaul to SupplyPike’s consumer demand modeling framework to improveaccuracy and reduce the costs of their retail demand forecasting models by 95%+.• Additionally worked on applications of AI/ML to automated data entry, collaborative filtering, and root cause analysis for inventory management.• Mitigated data quality problems associated with right-censored inventory data (stockouts).

Fayetteville, Arkansas, United States

• Collaborated with Walmart’s “2D to 3D” modeling team to construct tools decreasing Walmart’s reliance on costly physical samples from overseas suppliers in their design approval process.• Created a procedure for the nonparametric estimation of products’ raw material composition from manufacturing blueprints with incomplete information. This work leveraged Monte Carlo numerical integration methods from the GNU Scientific Library.• Presented code and its results to an audience of business leaders in McMillon Innovation Studio’s Demo Day 2020. • Additionally created, and presented, supplementary visualizations made using R and gnuplot.

Dayton, Ohio, United States

• Worked as a civilian contractor for the Air Force Research Lab-affiliated ATRC to engineer a novel technique for AI/ML ensemble learning with applications to sensors. This work was based on an original insight connecting the mathematics of ensembles to statistical control variates.• Led a team of three to create a prototype in PyTorch implementing the project’s design and experimentally test its performance on benchmark datasets.• First author of the associated research paper “Extending Negative Correlation Learning”, presented at a restricted conference and published to the ATRC’s internal database.

• Delivered a guest lecture on artificial neural networks and the unique challenges of Few Shot Learning to the Computer Science Department’s graduate-level Big Data elective.• Organized regular brainstorming and paper reading events within the lab.• Coauthored “Siamese Networks with Episodic Structure Loss for Few-Shot Learning”.