We introduce neural network layers based on operations on composite objects of scalars, vectors, and higher order objects such as bivectors. Published at ICLR 2023.
In this work, we introduce a message passing neural PDE solver that replaces all heuristically designed components in numerical PDE solvers with backprop-optimized neural function approximators. Published at ICLR 2022 (Spotlight).
We generalise steerable E(3) equivariant graph neural networks such that node and edge updates are able to leverage covariant information. Published at ICLR 2022 (Spotlight).
We generalize graph neural network based simulations of Lagrangian dynamics to complex boundaries as encountered in daily life engineering setups. Published at AAAI 2023.
We generalise steerable E(3) equivariant graph neural networks such that node and edge updates are able to leverage covariant information. Published at ICLR 2022 (Oral).
We introduce a modern Hopfield network with continuous states and a corresponding update rule. The new update rule is equivalent to the attention mechanism used in transformers. Published at ICLR 2021.