Baikal-GVD is a large-volume underwater neutrino telescope located in Lake Baikal, Russia. We report on machine learning techniques used for the analysis of its data. Namely, we discuss neural networks developed for the following goals: (1) suppression of noise activations of Baikal-GVD's optical modules due to the natural luminescence of Baikal water; (2) identification of neutrino-induced events and estimation of their flux; (3) reconstruction of arrival direction of incoming neutrinos. It is shown that the accuracy of developed methods surpass that of analogous standard reconstruction techniques on Monte-Carlo simulated data.