Speaker
Description
Modern Portfolio Optimisation faces computational bottlenecks. As the number of assets grow, the number of combinatorial solutions grow exponentially and become computationally difficult for classical computers to solve high dimensional problems from these large datasets, in the necessary time required. Financial computational strategies rely on the ability to acquire high-dimensional solutions as quickly
as possible. These strategies are easily out-performed by novel algorithms utilising High Performance Computing. Quantum Computing offers the potential to overcome these limitations, as they provide theoretical algorithmic speedups for NP-Hard problems. Real world Quantum Computers are inherently noisy. The arrival of the Utility era makes it pertinent to investigate whether they can solve real world problems let alone the theoretical computational speed-up of quantum methods over its classical counterparts. To investigate any possible utility, the work proposes two computational frameworks that solve the Global Mean Variance problem on JSE SWIX market data. A methodology is proposed to denoise Covariance Matrices as a solution to a Convex Program.
The denoised matrices constitute the datasets for both the U-Net Classical and Quantum Convolutional Neural Networks. The first framework is a U-Net Convolutional Neural Network to forecast a Covariance Matrix. A Convex Program solves the GMV on the new matrix. The second framework integrates a U-Net Quantum Convolutional Neural Network for forecasting and a Quantum Approximate Optimisation Algorithm to solve the GMV problem. We are yet to achieve viable results on Utility Quantum Computers.
| Topic | Hybrid quantum-classical approaches for learning and optimization |
|---|---|
| Additional Topic | Quantum variational circuits and their applications |