TINNs: Time-Induced Neural Networks for Solving Time-Dependent PDEs
Researchers have introduced Time-Induced Neural Networks (TINNs), a novel architecture designed to improve the solving of time-dependent partial differential equations (PDEs). Unlike traditional physics-informed neural networks (PINNs) that use a single network for all time steps, TINNs parameterize network weights as a function of time, allowing spatial representations to evolve dynamically. This approach, optimized with a Levenberg-Marquardt method, has demonstrated up to four times better relative error and ten times faster convergence in experiments compared to existing methods. AI
IMPACT This new architecture could lead to more efficient and accurate solutions for complex time-dependent problems in various scientific fields.