Speaker
Description
A Physics-Informed Neural Network (PINN) is a neural network that is constrained by laws of physics. The best-known type of PINN is a feedforward, fully connected neural network, or multi-layer perceptron, with a loss function that has a data term plus a term for the PDE that governs the physical system. Including physics knowledge that is additional to data reduces the solution space, which allows for finding a solution when limited data is available. A PINN is not necessarily a replacement for analytical or numerical methods; rather it is useful in cases where solutions are difficult to find with conventional methods. A PINN may also have a modified architecture of connections between neurons, but that is more difficult to do than informing the loss function. A PINN may be applied to finding a future state of a system given initial conditions, as is done in time-evolution simulations, and also for inverse problems in which the final state is known but the parameter values need to be determined. Examples will be presented.
Level for award;(Hons, MSc, PhD, N/A)?
N/A
| Apply to be considered for a student ; award (Yes / No)? | No |
|---|
