Pre-Accelerate Cambridge



Our technology is an engineering innovation comprised of a programmable circuit-level stochastic device combined with fast and efficient software transforms that can increase the speed of non-uniform random number generation multiple orders of magnitude beyond the current state of the art. Fast and efficient random number generation from non-uniform distributions is a significant bottleneck for a variety of calculations involving uncertain inputs and outcomes. Monte Carlo simulations, Bayesian inference, and generative adversarial neural networks all require a high-throughput source of samples from a non-uniform probability distributions. All three application areas are projected to grow rapidly as automation becomes increasingly prominent and computing systems are required to make increasingly more decisions based on uncertain information. Accelerators for non-uniform random generation currently do not exist. We are going to build one.