A recent paper by members of the DCIST alliance develops the use of learning techniques for the optimal design of wireless autonomous networks. This work observes that wireless optimization problems have a structure that is similar to statistical learning problems; however, one in which the statistical loss appears as a constraint. Stemming from this observation two natural ideas arise (i) The use of learning models to solve or approximate optimization problems in wireless autonomous networks in which the system loss functions are unknown or challenging to model. (ii) To conduct learning in the dual domain where constraints are linearly combined to create a weighted objective. The work explores the tradeoffs of learning in the dual domain and develops a gradient-based, primal-dual learning method. The framework is expanded with the introduction of a model-free learning approach, in which gradients are estimated by sampling the model functions and wireless channel. Tests utilizing deep neural networks show promise in simple scenarios and are currently being investigated for complex DCIST autonomous networking tasks.