Estimation of effective temperatures in quantum annealers for sampling applications: A case study with possible applications in deep learning

Abstract: An increase in the efficiency of sampling from Boltzmann distributions would have a significant impact on deep learning and other machine-learning applications. Recently, quantum annealers have been proposed as a potential candidate to speed up this task, but several limitations still bar these state-of-the-art technologies from being used effectively. One of the main limitations is that, while the device may indeed sample from a Boltzmann-like distribution, quantum dynamical arguments suggest it will do so with an {\it instance-dependent} effective temperature, different from its physical temperature. Unless this unknown temperature can be unveiled, it might not be possible to effectively use a quantum annealer for Boltzmann sampling. In this work, we propose a strategy to overcome this challenge with a simple effective-temperature estimation algorithm. We provide a systematic study assessing the impact of the effective temperatures in the learning of a special class of a restricted Boltzmann machine embedded on quantum hardware, which can serve as a building block for deep-learning architectures. We also provide a comparison to $k$-step contrastive divergence (CD-$k$) with $k$ up to 100. Although assuming a suitable fixed effective temperature also allows us to outperform one step contrastive divergence (CD-1), only when using an instance-dependent effective temperature do we find a performance close to that of CD-100 for the case studied here.

• The paper introduces a simple algorithm to accurately estimate instance-dependent effective temperatures in quantum annealers for optimized Boltzmann sampling.
• The study compares this method against CD-k, showing that its performance on restricted Boltzmann machines closely matches results akin to CD-100.
• Improved temperature estimation in quantum annealers offers enhanced deep learning training and broadens the practical applications of quantum-assisted AI.

Estimation of Effective Temperatures in Quantum Annealers for Sampling Applications

The paper, titled Estimation of Effective Temperatures in Quantum Annealers for Sampling Applications: A Case Study with Possible Applications in Deep Learning, provides an in-depth exploration of quantum annealers and their potential for sampling applications, particularly within the domain of deep learning. Quantum annealers are quantum computing devices designed to solve optimization problems by evolving quantum states towards the ground state of an Ising model. This paper primarily addresses the challenge of effective temperature estimation in these devices, which is crucial for their utilization in sampling from Boltzmann distributions.

A significant portion of the research investigates the limitations imposed by the instance-dependent effective temperatures of quantum annealers, as opposed to their physical temperatures. The differentiation between these two temperatures arises from the quantum dynamics involved in the annealing process, which leads to corrections in temperature that vary with different instances of the problem. Without accurately determining the effective temperature, the reliability of quantum annealers in sampling applications remains questionable.

The authors propose a simple algorithm to estimate these effective temperatures accurately. This estimation is essential for using quantum annealers to sample from Boltzmann distributions, which is a critical component in training certain machine learning models like restricted Boltzmann machines (RBMs) that can be foundational blocks for deep learning architectures.

The paper demonstrates the application of this temperature estimation method in training a class of RBMs on quantum hardware. A systematic study was performed to compare the performance of their approach against conventional methods, such as k-step contrastive divergence (CD-k), a popular method in training RBMs. Remarkably, they found that the use of an instance-dependent effective temperature often aligned the performance closely with CD-100, suggesting the quantum-assisted learning's efficacy in certain scenarios.

From a theoretical standpoint, the implications of this work are profound. The research supports the hypothesis that with accurate estimation, quantum annealers could effectively sample distributions more efficiently than classical counterparts. Practically, such advancements could lead to enhanced training processes for complex models, addressing limitations posed by classical approaches, especially for large-scale data and models with intricate data distributions.

The comprehensive analysis presented also points towards future developments in artificial intelligence where quantum computational resources are leveraged, promising improvements in the speed and accuracy of sampling, essential for learning more complex deep learning architectures.

While the study primarily utilizes the D-Wave 2X device, the insights and methodologies can be generalized to other quantum annealing systems. As the technology matures, it can be anticipated that these strategies will incorporate additional improvements in annealer design, control precision, and noise reduction, further increasing their applicability to various machine learning tasks.

In conclusion, the work contributes significant knowledge towards understanding and overcoming the challenges in quantum annealing for sampling applications, with implications extending to efficient training of deep learning models. This research serves as a foundational step towards utilizing quantum annealers more effectively in computationally intensive fields, potentially revolutionizing approaches in artificial intelligence and beyond.