Speaker
Description
This abstract is based on a work in progress that introduces a new type of neural field for visual computing with components compatible with gate-based quantum hardware or simulators. We propose a Quantum Neural Field Network (QNF-Net) expecting as input a query coordinate (of different dimensionality depending on the problem) and, optionally, a latent variable value. It outputs the corresponding learnt signal. QNF-Net includes a classical feature map providing quantum encoding of classical data and a quantum ansatz with parametrised quantum circuits. We observe in scenarios with 2D and 3D data that the QNF-Net trained on a simulator allows us to improve both the convergence speed and the representational accuracy (the PSNR metric emphasising high-frequency details) compared to strong classical MLP baselines.
