The prevalence of machine learning models in various industries has led to\ngrowing demands for model interpretability and for the ability to provide\nmeaningful recourse to users. For example, patients hoping to improve their\ndiagnoses or loan applicants seeking to increase their chances of approval.\nCounterfactuals can help in this regard by identifying input perturbations that\nwould result in more desirable prediction outcomes. Meaningful counterfactuals\nshould be able to achieve the desired outcome, but also be realistic,\nactionable, and efficient to compute. Current approaches achieve desired\noutcomes with moderate actionability but are severely limited in terms of\nrealism and latency. To tackle these limitations, we apply Generative\nAdversarial Nets (GANs) toward counterfactual search. We also introduce a novel\nResidual GAN (RGAN) that helps to improve counterfactual realism and\nactionability compared to regular GANs. The proposed CounteRGAN method utilizes\nan RGAN and a target classifier to produce counterfactuals capable of providing\nmeaningful recourse. Evaluations on two popular datasets highlight how the\nCounteRGAN is able to overcome the limitations of existing methods, including\nlatency improvements of >50x to >90,000x, making meaningful recourse available\nin real-time and applicable to a wide range of domains.\n