Jin and colleagues develop learning-based method for quantum sensing and networking

A thorough review of the increasingly intertwined fields of artificial intelligence and quantum information has been completed by Ming Cheng and colleagues at the University of Pittsburgh. This review details how quantum information presents new computational models and learning paradigms for AI development, while AI serves as a set of tools to advance the learning, design, control, and validation of quantum systems. This survey organizes recent advances on key tasks such as extracting information from limited measurements, training and discovering quantum algorithms, stabilizing hardware, automating workflows, and extending learning methods to sensing and networking. Additionally, this study examines the impact of quantum computation and quantum-inspired structures on learning while considering algorithmic speed, expressiveness, and neural network design, highlighting the critical need for integrated theoretical, experimental, and hybrid quantum-classical systems that enable overcoming challenges in reproducibility and scalability.

Using tensor networks to advance quantum learning and machine learning

Tensor network representations have proven to be central to enabling these advances, serving as a way to organize complex data into networks of interconnected nodes, similar to how a family tree shows relationships between individuals. Data represented in this interconnected format reduced the computational burden of high-dimensional information processing, a key challenge in both quantum simulations and advanced AI algorithms. These networks were used to model quantum states and complex connections within machine learning models, allowing for more efficient computation and analysis.

The fields of artificial intelligence and quantum information are becoming increasingly intertwined. Recent research details advances in using AI to improve quantum systems, focusing on tasks such as interpreting limited measurements and training quantum algorithms. The team also considered how quantum computing and related structures can enhance machine learning through improved algorithms and new ways to design neural networks. Further progress requires integrating theoretical developments with practical experiments and hybrid quantum-classical systems.

Achieving scalable quantum by mitigating barren plateaus through circuit design and tensor networks

Significant improvements in the training of quantum algorithms are detailed, increasing the success rate of parameter optimization by a factor of 10 compared to previous methods. Previously, optimization over a few qubits has proven difficult due to vanishing gradients and exponential scaling of resource requirements. Researchers at the University of Pittsburgh and NIST, in collaboration with multiple institutions across the United States, have found that careful consideration of circuit design and initialization strategies can alleviate the “barren plateau” phenomenon, a major obstacle in training complex quantum models.

Artificial intelligence (AI) and quantum information (QI) are rapidly coevolving, with AI becoming a practical tool for learning, designing, controlling, and validating quantum systems, and QI providing AI with new computational models and learning questions. Recent advances in AI for QI have focused on extracting information from limited measurements, training quantum algorithms, stabilizing noisy hardware, automating workflows, and extending learning methods to sensing and networking. Examining how quantum computation and quantum structures impact learning reveals the potential for faster algorithms, changes in expressive power, and new neural network designs. Tensor network representations provide a connection between quantum many-body structures, efficient representations, and practical machine learning applications. Advances in these fields depend on closer integration of theory, experiment, and hybrid quantum-classical systems.

Quantum benefits with stabilized systems and automatic parameter optimization

Combining artificial intelligence and quantum information science depends on overcoming practical limitations in both areas. While the potential of AI to stabilize quantum systems and improve algorithms is being explored, a crowded field of “parameter-efficient fine-tuning” techniques, including techniques such as LoRA and factor tuning, is already attempting to address similar challenges in classical machine learning. This raises important tensions over whether quantum-enhanced AI offers truly new benefits or simply replicates existing approaches, adding complexity and cost.

It is important to recognize the prevalence of parameter-efficient fine-tuning techniques in traditional machine learning. Researchers are exploring whether quantum systems can fundamentally improve these techniques, focusing on stabilizing quantum hardware and automating workflows to address critical bottlenecks that hinder practical quantum computing. Establishing the role of AI in overcoming these quantum challenges is a valuable step toward achieving truly new quantum-enhanced artificial intelligence, even if initial gains appear incremental.

Collaborative research between the University of Pittsburgh and partner institutions establishes a reciprocal relationship between artificial intelligence and quantum information science, where advances in one field directly inform and benefit the other. The team improved the training of quantum algorithms and achieved optimization of parameters that were previously limited by computational constraints and the “barren plateau” phenomenon, a major obstacle in developing complex quantum models. This success demonstrates the ability of AI to address practical challenges in building and controlling quantum systems, such as stabilizing delicate quantum states and automating complex experimental procedures.

This study demonstrated that artificial intelligence can successfully optimize the parameters of quantum algorithms and overcome limitations previously caused by computational constraints and the barren plateau phenomenon. This finding suggests that AI is a valuable tool for advancing practical quantum computing by helping to stabilize quantum systems and automate complex workflows. Researchers have achieved this by integrating AI technology and quantum information science, establishing a reciprocal relationship where advances in one field benefit the other. The authors emphasize the need for continued co-design of theoretical, experimental, and hybrid quantum-classical systems to further advance this integration.