Personalized computational modeling is rapidly expanding the landscape of therapeutic possibilities for neurostimulation. However, applications are limited by the computational expense of electrophysiological simulations. Here, we introduce a new method for dramatically reducing these costs while preserving the accuracy of traditional axon simulations.The “activating function” (AF) is a linear quantity derived directly from the extracellular field potential that can predict axonal spike initiation in lieu of compartment-based models [1], but with limited accuracy. Thus, we developed a “generalized activating function” (GAF) that extends the AF to account for conductive diffusion and leakage currents via convolution with a 1-D Green’s function.To test our methods, we conducted simulations in an anatomically detailed model of the spinal cord [2]. The electromagnetic (EM) fields arising from simulated epidural stimulation were computed in Sim4Life (ZMT Zurich MedTech AG, Switzerland), and associated axonal responses were predicted using the GAF. Stimulation parameters were optimized in PyTorch [3] for selectivity in a target spinal root. This procedure was repeated using detailed electrophysiological simulations. GAF- and electrophysiology-derived current thresholds for individual axons were in near perfect agreement. However, the GAF was five orders of magnitude faster.Techniques such as the AF offer significant (10^5 x) time savings compared to electrophysiological axon simulations, but at the cost of accuracy. In contrast, the GAF offers similar speed with excellent accuracy. In this way, the GAF unlocks exploration of large parameter spaces, with potential applications encompassing the optimization of complex stimulation configurations, pulse shapes, and electrode designs.