Karine Levonyan
Research Scientist / AI
I’m a research scientist who came to AI through computational physics and numerical linear algebra. That background shaped how I think: I want to understand models from first principles, not just use them.
At Stanford I spent years building intuition for how complex systems behave, and I bring that same curiosity to machine learning. These days I work on computer vision for face reconstruction, and graph neural networks for sparse data. Lately I’ve been thinking a lot about latent spaces. I think there’s something elegant about using them to peek inside neural networks and understand what they’ve actually learned.
Research Interests
- Machine Learning
- Computer Vision
- Facial Recognition
- Algorithmic Optimization
- Numerical Linear Algebra
- AI for Digital Content Creation
- Graph Neural Networks
- Embeddings for unstructured data
- Mechanistic Interpretability
- World Models
Recent Posts
- Solving the BlueDot Puzzle #1: XOR Geometry, DAS, and a Ring in the Velocity — June 24, 2026
- Internal Biases in the Visual Latent Space: A Comparative Empirical Study — May 18, 2026
- Building up to Navier–Stokes, One Term at a Time — April 08, 2026
Selected Publications
Towards ML-based Assessment of Synthetic Characters Heads I. Borovikov, K. Levonyan, P. Elliott, E. Danvoye IEEE 19th International Conference on Automatic Face and Gesture Recognition, 2025
System for Automated Generation of Facial Shapes for Virtual Character Models I. Borovikov, K. Levonyan, M. Anghelescu, D. Auclair, A. Ravikumar, et al. US Patent App. 18/194,074, 2024
Practical Parametric Synthesis of Realistic Pseudo-Random Face Shapes I. Borovikov, K. Levonyan, M. Anghelescu IEEE 17th International Conference on Automatic Face and Gesture Recognition, 2023
Puzzle Validation Using Heuristic Evaluation J.H.S. Harder, K.A. Levonyan, M. Sardari US Patent App. 17/643,106, 2022
Domain Engineering for Applied Monocular Reconstruction of Parametric Faces I. Borovikov, K. Levonyan, J. Rein, P. Wrotek, N. Victor arXiv preprint arXiv:2209.02600, 2022
Automated Graph Genetic Algorithm Based Puzzle Validation for Faster Game Design K. Levonyan, J. Harder, F.D.M. Silva IEEE Congress on Evolutionary Computation (CEC), 2022
Professional Service
- Reviewer, Empathic AI Workshop at IEEE FG 2026
Links
Education
Ph.D./MS in Computational Physics
Specializing in Energy Resources Engineering
Stanford University | 2019
Bachelor’s in Applied Mathematics
Gubkin University | 2008
Minor: Computer Science