Systematic comparison of deep belief network training using quantum annealing vs. classical techniques

Joshua Job,Steve Adachi

DOI: https://doi.org/10.48550/arXiv.2009.00134

2020-09-01

Abstract:In this work we revisit and expand on a 2015 study that used a D-Wave quantum annealer as a sampling engine to assist in the training of a Deep Neural Network. The original 2015 results were reproduced using more recent D-Wave hardware. We systematically compare this quantum-assisted training method to a wider range of classical techniques, including: Contrastive Divergence with a different choice of optimizer; Contrastive Divergence with an increased number of steps (CD-k); and Simulated Annealing (SA). We find that quantum-assisted training still outperforms the CD with Gibbs sampling-based techniques; however, SA is able to match the performance of quantum-assisted training trivially using a quench-like schedule with a single sweep at high temperature followed by one at the target temperature.

Quantum Physics

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is to evaluate the effectiveness of quantum annealing in training deep neural networks and make a systematic comparison with classical techniques. Specifically, the researchers re - examined and extended a 2015 study, which used the D - Wave quantum annealer as a sampling engine to assist in training the Deep Belief Network (DBN). The main objectives of the paper include:

1. **Verify previous results**: Confirm whether the results of the 2015 study can be reproduced on more modern D - Wave hardware.
2. **Systematic comparison**: Make a systematic comparison between the quantum - assisted training method and various classical techniques, including:
- **Contrastive Divergence (CD)**: Use different optimizers (such as Adam) and an increased number of steps (such as CD - 1000).
- **Simulated Annealing (SA)**: Use different cooling strategies, such as a quench - like schedule.

3. **Explore the source of performance advantages**: Determine whether the advantages of the quantum - assisted training method over classical methods are due to quantum effects or can be explained by improved classical techniques.

Through these studies, the paper aims to answer the following key questions:
- Is the performance advantage of the quantum - assisted training method really from quantum effects?
- Are there other classical techniques that can match or exceed the performance of quantum - assisted training?

Finally, the paper found that although the quantum - assisted training method is superior to the traditional contrastive divergence method in some cases, its advantages can be achieved by efficient classical algorithms (such as simulated annealing), thus indicating that current quantum annealing devices do not have a significant quantum advantage when training classical Restricted Boltzmann Machines (RBM).