Lyra Zhornyak
PhD Candidate · Dynamical Systems, Robotics & Physics-Guided ML
University of Pennsylvania · Advised by M. Ani Hsieh
I build physics-guided machine learning models to derive structure-preserving neural integrators from variational principles, predict critical transitions in dynamical systems, and optimize robot locomotion.
Research
Dynamical Systems
Currently building structure-preserving neural integrators that use variational principles to learn physically consistent simulations. Previously developed a transformer architecture that infers bifurcation diagrams from as few as 30 noisy trajectories (Chaos 2024).
Robotics
Optimized quadruped locomotion using central pattern generators, achieving smooth gait transitions with minimized joint torque. Found that optimal gaits closely match biological feline locomotion patterns, validating the approach for planetary rover designs (Robotica 2020).
Computer Vision
Demonstrated that Average Precision (AP) fails to penalize duplicate predictions in instance segmentation. Proposed Semantic Sorting + NMS, achieving 2x better detection accuracy, 33x fewer duplicates, and 6x faster inference (CVPR 2023).
Skills: Python, PyTorch, JAX, C/C++, MATLAB, Java, Haskell, Docker, Git, LaTeX
Publications
Inferring Bifurcation Diagrams with Transformers
Chaos: An Interdisciplinary Journal of Nonlinear Science, Vol. 34, No. 5, 051102
Beyond mAP: Towards Better Evaluation of Instance Segmentation
Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 21975–21984
HashEncoding: Autoencoding with Multiscale Coordinate Hashing
arXiv preprint, arXiv:2211.15894
Education
Ph.D. in Computer and Information Science (expected Spring 2027)
University of Pennsylvania · Advised by M. Ani Hsieh
Dissertation in progress on physics-guided machine learning for dynamical systems.
B.A.Sc. in Engineering Science
University of Toronto, Engineering Science · Supervised by M. Reza Emami
Thesis: Minimum torque gait transitions in quadruped rovers using central pattern generators.
Experience
Research
Graduate Research Assistant
GRASP Lab, University of Pennsylvania · Advised by M. Ani Hsieh
Designing and running computational experiments on nonlinear dynamical systems, building neural network pipelines for time-series prediction, and collaborating across robotics, control theory, and applied mathematics.
Industry
Engineering Intern
Orbis Investment Management, Data Science Team · London, UK
Designed and built a computer vision pipeline for real-time brand logo detection in video streams to support investment research analysis.
Undergraduate Technical Intern
Intel, Programmable Solutions Group · San Jose, CA
Built a constraint verification system for PCIe module configuration on the Intel Agilex FPGA platform. Implemented message-signaled interrupt (MSI) support in Linux kernel drivers.
Teaching & Service
Peer Reviewer
Chaos: An Interdisciplinary Journal of Nonlinear Science; IEEE Robotics and Automation Letters (RA-L); Chaos, Solitons & Fractals
Reviewing manuscripts on nonlinear dynamics, robotic systems, and chaos theory for leading journals in the field.
Teaching Assistant
University of Pennsylvania · CIS 680: Advanced Topics in Machine Perception
Graduate-level course. Held office hours, graded assignments, and supported students across multiple semesters.
Control Team Lead
University of Toronto Humanoid RoboSoccer Team
Led the control subsystem for autonomous humanoid robots competing in international RoboCup events. Responsible for locomotion, balance, and behavior planning.
Awards
Mitacs Globalink Research Award
University of Leeds, UK
Competitive international research fellowship funding collaboration abroad. Project: Vision-based tracking of hands in vehicles.
NSERC Undergraduate Student Research Award
University of Toronto
National peer-reviewed research grant from Canada's federal science funding agency. Project: Mobility analysis and synthesis of quadruped robots.
Contact
I am open to research collaborations and industry or postdoctoral opportunities. Reach me at zhornyak@seas.upenn.edu.