High-Dimensional Signal Processing using Classical-Quantum Machine Learning Pipelines with the TensorFlow Stack, Cirq-NISQ, and Vertica

Modeling High Dimensional Signal Data (HDSD) is challenging due to statistical complexities and an imbalance of noise to signal data, which creates ground truth problems. Likewise, enabling recognition and detection analytics on the information-bearing portion of the noise-obscured HDSD is computationally cost-prohibitive due to the unprecedented size, speed, and scale of HDSD. Consequently, human experts often out-perform Deep Learning (DL) algorithms at HDSD recognition. This is problematic due to the spectrum of applications reliant on accurate HDSD detection. These include multi-sensor and biologic signal detection, virus outbreak surveillance, genetic association identification, image, audio, and sonar processing, source separation for speech identification, astrophysical source detection, financial time series, and satellite payload processing. To solve this problem and facilitate favorable DL outcomes, synthetic data possessing enough positive samples is needed to balance the HDSD dataset. Research reveals an entangled Quantum Generative Adversarial Network (QGAN) utilizing parameterized quantum circuits for the QGAN Generator (QGEN) and Discriminator (QDIS) exhibit an exponential advantage over a classical-GAN when generating highly accurate and cost-effective synthetic data. Using this method, an unsupervised TensorFlow-Quantum QGAN is prepared from Cirq Noisy Intermediate-Scale Quantum (NISQ) circuits. The QGEN and QDIS create new statistically consistent synthetic HDSD from real data; thereby solving the ground truth problem. Next, TensorFlow-GPU facilitates classical-DL to train the new synthetic HDSD dataset. Once validated, the TensorFlow model is exported to the production Vertica analytics tier for fast and accurate ML inference, where the HDSD is instantly identified and classified at ingest.