Abstract
Variational quantum machine learning is an extensively studied application ofnear-term quantum computers. The success of variational quantum learning modelscrucially depends on finding a suitable parametrization of the model thatencodes an inductive bias relevant to the learning task. However, preciouslittle is known about guiding principles for the construction of suitableparametrizations. In this work, we holistically explore when and how symmetriesof the learning problem can be exploited to construct quantum learning modelswith outcomes invariant under the symmetry of the learning task. Building ontools from representation theory, we show how a standard gateset can betransformed into an equivariant gateset that respects the symmetries of theproblem at hand through a process of gate symmetrization. We benchmark theproposed methods on two toy problems that feature a non-trivial symmetry andobserve a substantial increase in generalization performance. As our tools canalso be applied in a straightforward way to other variational problems withsymmetric structure, we show how equivariant gatesets can be used invariational quantum eigensolvers.