Artificial intelligence and machine learning have drastically changed society and technology in the past decade, with the potential for further disruption in the near future. Quantum computing, in particular, has become a topic of interest for government entities, large corporations, and academics. Quantum machine learning, which combines machine learning algorithms and quantum processors, has successfully allowed scientists to better control quantum systems. However, the field of quantum biomimetics seeks to bridge the gap between natural biological systems and quantum devices by replicating biological behaviors in quantum-controllable systems. The ultimate goal is to design more efficient artificial devices. This special issue showcases the interplay between quantum machine learning and quantum biomimetics and features articles focused on quantum machine learning, quantum biomimetics, and their fusion.

The hybrid quantum-classical framework has become increasingly popular amongst researchers in both classical machine learning and quantum physics and chemistry. While current models have only been implemented on small to intermediate scale systems or datasets, they provide insight into the potential large-scale applications that may arise in the future. The hybrid framework is comprised of a parametrized quantum circuit and classical optimizers which can significantly impact the performance of the model. This paper compares the performance of nine widely used optimizers, including gradient and gradient-free techniques, and evaluates their effectiveness in a supervised learning scenario. The results showed that gradient-based optimizers offer better solutions in each case, but gradient-free techniques can be more practical for smaller problems due to their lower running time. The findings from this study may be useful for researchers interested in similar problems.

Large-scale experiments in particle physics rely heavily on advanced information processing techniques to store, process, and analyze the vast amounts of data they produce. The High-Energy Physics (HEP) field has benefited from advancements in information processing and the development of algorithms and tools for large datasets. The field has also explored quantum computing applications to understand how the community can benefit from the advantages of quantum information science. However, there is a need to understand the quantum behavior and scale up current algorithms to unleash the full potential of quantum computing. In this context, this work explores potential applications of quantum machine learning to data analysis tasks in HEP and how to overcome the limitations of algorithms targeted for Noisy Intermediate-Scale Quantum (NISQ) devices. By leveraging quantum machine learning, physicists could potentially achieve faster and more accurate data analysis, leading to further scientific breakthroughs in the field.

Artificial neural networks can be vulnerable to malicious inputs, known as adversarial attacks. However, the extent of these vulnerabilities in the quantum machine learning (ML) setting is not fully understood. Researchers have benchmarked the robustness of quantum ML networks, such as quantum variational classifiers (QVC), against classical adversarial attacks. They found that QVCs offer superior robustness against classical adversarial attacks by learning features not detected by classical neural networks, indicating a possible quantum advantage for ML tasks. The converse is not true, with attacks on quantum networks also capable of deceiving classical neural networks. The researchers propose a novel adversarial attack detection technology by combining quantum and classical network outcomes. The study reveals the potential for a new kind of quantum advantage through superior robustness of ML models, addressing serious security concerns and reliability issues of ML algorithms used in applications such as autonomous vehicles, cybersecurity, and surveillance robotic systems. This finding expands the traditional notion of quantum advantage beyond increased accuracy or algorithmic speed-up.

Researchers have developed a new technique, called Quantum Shift Based Recurrent Neural Machine Learning (QS-RNML), for efficient gaze pattern recognition, which predicts where the human eyes are pointing in a predefined plane. The new technique includes three processes: preprocessing, gaze estimation, and pattern recognition. First, the adaptive median filter is employed to remove noise artifacts from eye images in the preprocessing stage. Then, Quantum Shift is applied to estimate the patterns. Finally, recurrent neural machine learning is used to recognize the gaze patterns by matching the estimated patterns with the ground truth patterns. By implementing the QS-RNML technique, researchers observed improvements in gaze pattern recognition accuracy, true positive rate, while reducing time complexity and false positive rate. The proposed method improves the accuracy of human visual attention, emotions, and feelings detection. This technique is both simple and efficient and could find many applications in various interactive systems.

Researchers have explored the quantum chemical foundations of descriptors for molecular similarity, which are important for machine learning to traverse chemical compound space. The focus of the study was on the Coulomb matrix and the smooth overlap of atomic positions (SOAP). The researchers have defined two new descriptors more closely related to electronic structure theory, Coulomb lists and smooth overlap of electron densities (SOED), and gained insight into how and why Coulomb matrix and SOAP work. Coulomb lists can extract subsystem information that can be compared across Born-Oppenheimer surfaces of varying dimension. The researchers derived the formalism to create the SOED measure in close analogy to SOAP and introduced approximations to work with SOED. The study focused on elementary reaction steps, where transition state structures are more similar to either reactant or product structures than the latter two are with respect to one another. The results raise the question of the extent to which molecular similarity descriptors rooted in electronic structure theory can resolve multi-configurational effects in the prediction of electronic energies of transition state structures.