I discuss my current research in using contour deformations to alleviate the numerical sign problem in stochastic simulations. I consider deformations of various forms, ranging from simple constant offsets to those that approximate so-called Lefschetz thimbles--high-dimensional manifolds that have, in principle, no sign problem. In the latter case I show how machine learning can be used to approximate such manifolds. I apply these deformations to investigate low-D doped Hubbard model in various geometries.