Radmir S | Vitae

Experience

Proficient with Python, Matlab, PyTorch, TensorFlow, OpenCV, MMlab and other ML frameworks
Highly experienced in registration, classification and segmentation of medical images (nifti/dicom)
Achieved 97.5% accuracy on cardiac diagnosis problem using Diffeomorphic Registration and Random Forest
Implemented DL models as VoxNet, PointNet, Autoencoders for analysis of 3D MRI images
Performed image processing tasks, including coarsening, refinement, inpainting, PCA alignment, ICP registration.

Houston, TX
2021 - present

Built an API for face detection and face tracking application using MMpose and FaceNet toolboxes
Automated and standartized ML model retraining pipeline on GCP/VertexAI cloud machine
Tested and integrated pose and object detection models, such as ResNet, YOLOv, and TCFormer, into the application
Built and trained custom NN model for physical parameter estimation with 98\% hold-out accuracy using PyTorch

Houston, TX
Summer 2022

Implemented and fine-tuned quadratic interpolation for Delta/Rho variables producing 3-5% rel.error of approximation
Worked on solutions of reducing the computational cost of the Greeks estimation for intra-day options trading

Minneapolis, MN
Summer 2020

Education

Ph.D. in Applied Mathematics

Houston, TX
2020 - 2024

M.S. in Mathematics

Detroit, MI
2018 - 2020

B.S. in Mathematics

Almaty, Kazakhstan
2005 - 2009

Papers

Skills & Knowledge

Programming: Python, MATLAB, C++, R, SQL
Frameworks: PyTorch, TensorFlow/Keras, opencv, pandas, sci-kit, spark, sql, git, SLURM, ssh/bash, GCP, docker, vscode remote, omp, multiprocessing, ANTs, labelme, slicer
Relevant coursework: Optimization, Probability & Statistics, Numerical Methods, Deep Learning, Data-Driven Algorithms, Statistical Data Analysis, High-Performance Computing, Linux/Cluster Computing
Certifications: TensorFlow Developer Certificate, Transfer Learning for Images Using PyTorch, Linux, etc

Machine Learning Course Projects

Age Recognition | Data-Driven Algorithms, University of Houston
I conducted age recognition problem as my final project during the Pattern Recognition course I completed at the University of Houston. I transformed 30k+ of face images from Kaggle into 128 measurements using OpenFace. Then I implemented the PCA analysis and ML algorithms such as SVM, Random Forest to solve the age recognition problem. Repo.
Financial Hedging | Math-to-Industry Boot Camp, Securian Financial
We worked in a team researching solutions for reducing the computational cost of the estimation of the Greek variables on the options market. My main contribution was to test performance of quadratic interpolation as a proxy model for the Black-Scholes model for Delta and Rho variables producing 3−5% relative error of approximation. Repo.
Match Prediction | Statistical Data Analysis, Wayne State University
As part of the Statistical Data Analysis course at Wayne State University, I conducted a final project focused on match prediction problem. I have collected 5k+ tennis match data from internet websites using parsing Python packages. Then I implemented ML algorithms such as KNN, QDA, LDA, Ridge&Lasso methods producing 89% prediction with KNN. Repo.