PROJECT SYNOPSIS
We propose a novel method for using Quantum Annealing Computers (QACs) to train and sample from Deep Boltzmann Machine (DBM) model distributions, combining QACs with classical sampling techniques. The search space for optimal parameters in complex deep learning (DL) systems is vast and conventional algorithms are often computationally expensive, leading to suboptimal solutions. Quantum computing (QC) offers the potential for faster training and better parameter discovery within these large search spaces. Our goal is to demonstrate that QC-based training can outperform traditional optimization methods, effectively addressing limitations in current DL technologies, particularly for challenges like automatic interpretation of electroencephalography (EEG) signals.
IMPACT
We aim to enhance undirected generative machine learning model's
training by leveraging Quantum Annealing Computers (QACs) to better
handle complex, unbalanced datasets. By combining QAC with classical
sampling, we want to boost trainability and identify high-probability
states often missed by traditional methods. This approach promises
significant improvements in rare event classification, with potential
applications in areas like digital pathology and EEG signal
interpretation.
The data and technology being developed in this project is made
available as open source resources. Please go
here
to learn more about the data and corpora available.
