University of Pittsburgh
PhD Researcher
Signal Processing and Statistical Learning Lab (SPSL)
Aug 2025–present
Ege Ozkoc
PhD Student, Electrical and Computer Engineering
University of Pittsburgh
My research focuses on machine learning and signal processing for real-world time-series data, with current work on interpretable AI-ECG systems and LLM-based explanations. My broader work spans efficient deep learning and practical ML systems, from model compression and edge inference to low-level implementation.
Email · Resume · Google Scholar · GitHub · LinkedIn
Selected Work
Interpretable AI-ECG with LLMs
I develop evidence-grounded LLM pipelines for interpreting AI-ECG predictions, combining waveform evidence, model outputs, and saliency/SHAP attributions. This work has led to a U.S. provisional patent application.
Real-Time Parkinson's Tremor Detection
At Fraunhofer IIS, I developed a lightweight 1D CNN for Parkinson's tremor detection using wrist-worn IMU data. It achieved approximately 0.91 AUC on unseen subjects, while 8-bit quantization reduced model size by approximately 75% with negligible performance loss.
Bayesian Inverse ECG Reconstruction
I used Bayesian MAP estimation to reconstruct cardiac electrical activity from body-surface measurements, evaluating prior models derived from measured and simulated data. This work resulted in two peer-reviewed conference publications.
ML Systems & Open Source
I build ML systems and implementations beyond high-level model APIs, including a deep-learning framework from scratch, a C++ array library, offline Whisper dictation using MLX, and merged bitsandbytes contributions.
deep-learning-from-scratch · mini-numpy · murmur · bitsandbytes contributions
Additional Projects
- lora-from-scratch — a minimal PyTorch implementation of LoRA, including low-rank adapters, injection into existing linear layers, weight merging, and a small LLM fine-tuning experiment.
- clustering-toolbox — a Streamlit application for clustering and exploring tabular data with KMeans, Gaussian mixture models, DBSCAN, PCA, and t-SNE.
Publications
- Ege Ozkoc, Tobias Sebastian Zech, Norman Pfeiffer, Stephan Göbl, and Jürgen Frickel. “Compressed and Lightweight CNN for Real-Time Parkinson’s Tremor Detection from Wearable IMU Data.” 2025 IEEE International Workshop on Machine Learning for Signal Processing (MLSP), Istanbul, Türkiye, 2025. Paper
- Ege Ozkoc and Yesim Serinagaoglu Dogrusoz. “Bayesian MAP Solution of the Inverse ECG Problem with Sinus Rhythm Data: Evaluation of Simulated Training Sets.” 30th IEEE Signal Processing and Communications Applications Conference (SIU), 2022. Paper
- Ege Ozkoc, Elifnur Sunger, Kutay Ugurlu, and Yesim Serinagaoglu Dogrusoz. “Prior Model Selection in Bayesian MAP Estimation-Based ECG Reconstruction.” 13th International Conference on Measurement, 2021. Paper
Experience
Fraunhofer Institute for Integrated Circuits IIS
Master's Thesis Researcher
Medical Sensors and Analytics Group
Mar 2024–Nov 2024
FAU Erlangen-Nürnberg
Research Intern & Student Assistant
Machine Learning and Data Analytics Lab
Jun 2023–Apr 2024
Universitätsklinikum Erlangen
Student Research Assistant
Department of Radiology
Feb 2023–Oct 2023
Middle East Technical University
Undergraduate Researcher
Heart Research Laboratory
Jul 2020–Jul 2022
ASELSAN
Part-time Candidate Engineer
Radar & Electronic Warfare · Nov 2021–Jan 2022
Engineering Intern
Defense Systems Technologies · Jul 2021–Aug 2021
2021–2022
Education
University of Pittsburgh
PhD in Electrical and Computer Engineering
2025–present
FAU Erlangen-Nürnberg
MSc in Medical Engineering
2022–2024
Middle East Technical University
BSc in Electrical and Electronics Engineering
2016–2022