13:00 - 13:30 Toby van Gastelen (Scientific Computing), Energy-conserving neural networks for turbulence modeling
Abstract: In this talk I will discuss turbulence modeling and its challenges. I will go in to how we can harness the power of neural networks to obtain better turbulence models. In addition, I will discuss the problems with these neural networks. Namely, they are not aware of the physical conservation laws. This can cause instabilities in the resulting turbulence simulation. Finally, I will go in to how we to use the concept of energy conservation to resolve these issues.
13:30 - 14:00 Aditya Gilra (Machine Learning), Deriving biologically-plausible learning rules for brain-like neural networks via adaptive control theory
Abstract: Artificial neural networks are inspired by neuronal networks in the brain, but they are typically trained using backpropagation of error. Backpropagation uses quantities that are not easily available locally at the synaptic connections between neurons, making it an implausible mechanism for learning in the brain. Experimentally, various local synaptic plasticity rules have been discovered between biological neurons, but how these can produce learning for the organism's behaviour is still unclear. I will present how biologically-plausible local synaptic plasticity rules can be derived from adaptive control theory, to enable spiking recurrently-connected neuronal networks to learn to predict and control nonlinear body dynamics.