Davit
Shadunts
ML Engineer | Software Engineer | Researcher
ML Engineer | Software Engineer | Researcher
Everything you see across this site, from fractals to flowing patterns, has been coded entirely from scratch.
Make sure to visit my portfolio for more examples of exploring mathematical concepts through code.
With 5 years of experience in industry and advanced research at Columbia University, I bring together engineering and research, specializing in machine learning, signal processing and the application of foundation models to vision and control.
This project enhances Meta’s DINO World Model by integrating a bisimulation loss signal into the vision representation model DINOv2.
The bisimulation distance encourages latent states to capture task-relevant behavioral similarity, promoting sample-efficient control and better generalization.
This work is pending publication and was conducted under the supervision of Professor James Anderson at Columbia University.
