Adiabatic Quantum Computation Applied to Deep Learning Networks.

Area of Science:

• Quantum Computing
• Artificial Intelligence
• Machine Learning

Background:

• Deep learning network training is computationally intensive, often requiring simplified topologies for graphical processing units (GPUs).
• Quantum devices offer new possibilities for training complex network topologies.
• The D-Wave processor utilizes quantum annealing, a restricted form of adiabatic quantum computation.

Purpose of the Study:

• To explore quantum annealing for training complex deep learning network topologies.
• To classify MNIST data and neutrino detection data using quantum annealing.
• To investigate performance improvements using expanded topology options on quantum processors.

Main Methods:

• Translating image and neutrino detection data into Ising models for quantum annealing.
• Utilizing a D-Wave quantum processor to train a specific network topology.
• Comparing quantum annealing with high-performance computing (HPC) and neuromorphic computing approaches.

Main Results:

• Quantum annealing successfully trained complex network topologies for data classification in reasonable time.
• High-performance computing identified good parameters for simplified network topologies.
• Neuromorphic computers demonstrated low-power solutions for complex topologies using memristive hardware.

Conclusions:

• Quantum annealing is a promising approach for training complex deep learning networks, overcoming traditional computational limitations.
• Alternative methods like HPC and neuromorphic computing offer complementary solutions for deep learning network training.
• The study highlights diverse strategies for advancing deep learning training methodologies.