🎉 2 Papers accepted to AAAI 2025! 🎉
We are happy to announce that the lab has two papers accepted to AAAI 2025! Congratulations to our students, Elifnur Sunger and Paul Ghanem, for their incredible works:
- ⭐
MarkovType: A Markov Decision Process Strategy for Non-invasive Brain-Computer Interfaces Typing Systems
- ⭐
Learning Physics Informed Neural ODEs With Partial Measurements
MarkovType: A Markov Decision Process Strategy for Non-invasive Brain-Computer Interfaces Typing Systems
Learning Physics Informed Neural ODEs With Partial Measurements
MarkovType: A Markov Decision Process Strategy for Non-invasive Brain-Computer Interface Typing Systems
Elif proposed a novel method for the Rapid Serial Visual Presentation (RSVP) paradigm in Brain-Computer Interfaces (BCIs) by framing the recursive classification problem of the typing task as a Partially Observable Markov Decision Process (POMDP). The proposed method, MarkovType, uses reinforcement learning with a policy gradient approach to optimally select the query characters to display via RSVP to the user. By integrating the typing mechanism into the learning process, MarkovType enhances both accuracy and the balance between speed and precision, resulting in a significantly superior typing system compared to competing methods.
Physics Informed Neural ODEs With Partial Measurements
Paul proposed a sequential optimization framework inspired by state estimation theory and Physics-Informed Neural ODEs to model the dynamics of physical and spatiotemporal processes with partially unmeasured states. Specifically, his method handles systems where the dynamics of these unmeasured states are unknown. He showcased the framework's superior performance on numerical simulations and a real-world dataset derived from an electro-mechanical positioning system.