Large-scale online learning of implied volatilities

Online-learning-based approaches do not suffer from generalization errors because the data are used once anddiscarded rather than reused. This characteristic enables incredible accuracy in estimating implied volatilitiesover a wide input area, outperforming existing state-of-the-art studies. In addition, an iterative method furtherimproves the estimates. The iterative method takes the estimate as an initial guess and corrects it, quicklygiving a virtually true value. Additionally, optimizing the network structure with TensorRT significantlyreduces the time spent on these processes