Neurosymbolic systems
In a neurosymbolic system, a neural network turns raw input into symbols and a symbolic component (an automaton, a logic program or a probabilistic reasoner) draws conclusions from them. We study how perturbations of the neural part propagate to the conclusions of the whole system: scalable robustness verification of probabilistic neuro-symbolic reasoning, end-to-end verification of neuro-symbolic automata, and the strategic abilities of agents built from neural and symbolic components.


