A significant challenge in ML-based Physics Informed Neural Networks (PINNs) is their limited generalizability. Traditional PINNs often require retraining for different conditions within the same geometry, hindering their practical application. Our team at BosonQ Psi (BQP) has developed a novel approach by introducing a quantum layer to PINNs. This research, led by Rut Lineswala, Abhishek Chopra, and Jay Shah, has significantly enhanced model accuracy while reducing computational costs. By overcoming the generalizability challenge, our QA-PINN model opens up new possibilities for rapid design exploration and optimization across various industries. We are excited to present our research at the IEEE Quantum Week in Montreal. #CFD #QuantumComputing #AI #ML #PINN #Research #Innovation #Science #Engineering #IEEE #QuantumWeek
Grateful and excited to see our work at BosonQ Psi (BQP) on Quantum-Assisted Physics Informed Neural Networks (QA-PINN) for #CFD get accepted to be showcased at the prestigious #IEEE #Quantum Week in Montreal in the Fall. Special shoutout to Jay Shah for his exceptional contribution. Rut Lineswala and I are very proud to see our work, which falls in the intersection of #engineering, quantum, and #AI/#ML, being recognized by highly technical and distinguished reviewers. Problem: ML-based frameworks, especially, #PINN though interesting suffer from the issue of generalizability - how can you test for multiple conditions for the same geometry without retraining the entire model? Solution: In this study, we proposed that by adding a quantum layer, we can reduce the training parameters while maintaining (and even) improving the accuracy. Outlook: This is a first step towards a larger discovery. ML-based CFD approaches are not an alternative to high-fidelity CFD approaches, but rather a design study tool to help engineers test a larger design space quickly and focus on the critical test conditions with high-fidelity CFD models.