This project implements a Differential Neural Network (DNN) in C++. It was developed as part of my M.Sc. in High-Performance Computing Final Project. The Differential Neural Network concept is inspired by the work of Antoine Savine, as detailed in this paper.
The Differential Neural Network is a model designed for high-performance and computationally efficient learning, inspired by techniques in automatic differentiation and gradient-based optimization. The model utilizes a Twin-Network architecture, essentially comprising two Neural Networks working in tandem to output a predicted value alongside predicted derivatives of inputs for a given set of inputs. My implementation showcases this process with randomly generated Black-Scholes Data.
- Parallel C++ Implementation: The code is optimized for high-speed execution and efficient memory usage.
- Twin-Network Architecture: Utilizes a dual-network setup to enable differential calculations with enhanced precision.