Opportunities and challenges for quantum-assisted machine learning in near-term quantum computers (1708.09757v2)

Abstract: With quantum computing technologies nearing the era of commercialization and quantum supremacy, ML appears as one of the promising "killer" applications. Despite significant effort, there has been a disconnect between most quantum ML proposals, the needs of ML practitioners, and the capabilities of near-term quantum devices to demonstrate quantum enhancement in the near future. In this contribution to the focus collection on "What would you do with 1000 qubits?", we provide concrete examples of intractable ML tasks that could be enhanced with near-term devices. We argue that to reach this target, the focus should be on areas where ML researchers are struggling, such as generative models in unsupervised and semi-supervised learning, instead of the popular and more tractable supervised learning techniques. We also highlight the case of classical datasets with potential quantum-like statistical correlations where quantum models could be more suitable. We focus on hybrid quantum-classical approaches and illustrate some of the key challenges we foresee for near-term implementations. Finally, we introduce the quantum-assisted Helmholtz machine (QAHM), an attempt to use near-term quantum devices to tackle high-dimensional datasets of continuous variables. Instead of using quantum computers to assist deep learning, as previous approaches do, the QAHM uses deep learning to extract a low-dimensional binary representation of data, suitable for relatively small quantum processors which can assist the training of an unsupervised generative model. Although we illustrate this concept on a quantum annealer, other quantum platforms could benefit as well from this hybrid quantum-classical framework.

Quantum-Assisted Machine Learning in Near-Term Quantum Computers: A Focus on Opportunities and Challenges

The rapid advancement toward quantum computing's commercial availability and quantum supremacy has positioned quantum-assisted machine learning (QAML) as a potential significant area of application. This paper discusses the various prospects and hurdles associated with harnessing near-term quantum computers to enhance ML tasks, especially those classified as intractable by current classical computing standards.

Overview of Opportunities for QAML

1. Focus on Intractable ML Tasks: The paper identifies generative models within unsupervised and semi-supervised learning as key opportunities for QAML. Unlike supervised learning, these models are computationally intensive and have not been fully addressed by classical methods. Quantum devices could potentially offer efficient sampling from complex probability distributions, accelerating inference and learning in these models.
2. Datasets with Quantum-Like Correlations: The authors propose using quantum computers to model datasets with inherent quantum characteristics. Examples from cognitive science suggest certain human behavior datasets exhibit non-classical probability patterns, potentially giving quantum models an advantage over classical ones in these contexts.
3. Hybrid Quantum-Classical Architecture: For near-term practicality, the paper emphasizes hybrid architectures where quantum devices are employed for specific intractable tasks within the classical ML pipeline. This approach leverages quantum efficiency in handling mathematical intractability while benefiting from classical computing's familiarity and robustness in broader tasks.

Challenges to Implementation

1. Model Compatibility: A significant challenge lies in achieving coherence between quantum and classical components of a hybrid ML algorithm—particularly concerning temperature estimates and sampling distributions.
2. Robustness to Noise: Quantum devices often experience noise in programmable parameters, potentially leading to deviations from the desired Gibbs distribution. Building gray-box models that operate with noisy data or estimating effective temperatures can mitigate this issue.
3. Connectivity Constraints: Quantum devices have limited qubit connectivity, impacting model topology and requiring creative embedding strategies or significant computational overhead to approximate desired logical connections.
4. Complex ML Dataset Representation: Near-term devices have limitations in handling large, high-dimensional datasets. Strategies such as semantic binarization, where data are stochastically mapped to abstract binary representations, may pave the way for feasible implementations on limited-qubit quantum computers.

Theoretical and Practical Implications

Practically, QAML indicates high potential in fields where classical methods fall short, such as unsupervised learning. This potential can inspire new approaches for data-driven sciences, offering an intersection between quantum computing and applied statistics. Theoretically, insights gained from exploring quantum properties in datasets could expand understanding beyond traditional machine learning frameworks, particularly in unique domains like cognitive sciences.

Future Directions

Forward-thinking involves developing custom quantum architectures tailored to specific ML tasks, exploring hybrid algorithms capable of effectively leveraging both classical and quantum resources, and identifying or creating datasets with distinct quantum correlation structures that can be efficiently modeled by today's emerging quantum technologies. Exploration towards quantum Gibbs distributions and other specific quantum-stochastic representations is slated as a domain of particular interest for near-future research.

In conclusion, despite the technological and conceptual challenges, the promise of QAML in transforming complex, hard-to-solve ML tasks remains substantial. Continued multidisciplinary efforts are imperative for unlocking the full capabilities of quantum computers in the field of machine learning.