- Worked on IBM Bob, an enterprise AI agent coding platform, extending agentic developer workflows with native session management and real-time agent interaction.
- Built predictive models for token consumption and spend within Bobalytics, Bob’s usage analytics dashboard, delivering cost prediction and team-level adoption metrics for enterprise budget planning.
Ph.D. Student • Machine Learning • Data Analytics
Onur
Vural
I build machine learning models that read the Sun's magnetic field as multivariate time series and predict the flares that follow.
Deep learning, time series mining, and classification under extreme class imbalance.
Previously & presently
Utah State University
IBM
NASA
Northrop Grumman
Bilkent University
Background
About
Hello! I am Onur Vural, a Ph.D. student and research assistant with expertise in machine learning, deep learning, and data analytics, particularly in the field of solar physics. My primary focus is on leveraging advanced machine learning techniques to analyze and predict solar phenomena, such as solar flares, using multivariate time series data.
My research spans representation learning, time series mining, and the development of novel algorithms for imbalanced datasets. I am passionate about using innovative computational methods to extract meaningful insights from complex datasets, enabling advancements in both science and technology.
When I'm not immersed in research, I enjoy collaborating on projects that push the boundaries of artificial intelligence and sharing my knowledge through talks, videos, and publications. Feel free to explore this website to learn more about my work, ongoing projects, and other contributions.
Industry & national labs
Experience
- Worked on the world’s first neural twin framework integrating physics AI models within a simulation environment for AI-enriched aerospace design and validation. Built the test harness for parallel AI surrogate model execution in simulation.
- Contributed to NASA’s AiTHENA initiative by interjecting AI/ML solutions into a unique, agile, and cost-efficient workflow across the agency to enhance data-driven scientific research and decision-making.
- Demonstrated feasibility of state-of-the-art time series forecasting techniques to improve input data quality and real-time reliability of the Nowcast of Atmospheric Ionizing Radiation System (NAIRAS) model.
- Developed an integrated data-driven pipeline and machine learning evaluation framework to support scalable processing of air traffic simulation outputs and to automate air traffic management scenario analysis while maintaining interpretability.
Published work
Research
Global Cross-Time Attention Fusion for Enhanced Solar Flare Prediction from Multivariate Time Series
A transformer that injects learnable global tokens, cross-attending over the full sequence to surface the non-contiguous moments that actually carry flare signal.
Solar Flare Prediction Using Multivariate Time Series of Photospheric Magnetic Field Parameters
A comparative analysis of vector, time series, and graph formulations of the same magnetic field data, measured against a shared benchmark.
EXCON: Extreme Instance-based Contrastive Representation Learning of Severely Imbalanced Multivariate Time Series
Selects the most distinctive instance per class, then trains a temporal embedding module with a custom extreme reconstruction loss. Transfers to univariate and multiclass settings.
Contrastive Representation Learning for Predicting Solar Flares from Extremely Imbalanced Multivariate Time Series Data
Contrastive objectives applied where the positive class is vanishingly rare, pulling like instances together and pushing the rest apart in the embedding space.
Talks
Videos
Introduction to Solar Data Analytics
What solar observatories actually record, and how those recordings become a dataset.
Machine Learning for Multivariate Time Series
Methods for classifying long, noisy, heavily imbalanced sequences.
All videos
Talks, walkthroughs, and paper explanations.
Press
News
Here Comes the Sun: USU Computer Scientists Develop Models to Predict Extreme Solar Phenomena
University coverage of the group's work on predicting extreme solar events.
Recent activity
Conference talks, paper acceptances, and shorter updates.