Dr Shuxiang Cao

Abstract:

Fully automated self-driving laboratories promise high-throughput, large-scale scientific discovery by reducing repetitive labor. However, they require deep integration of laboratory knowledge, which is often unstructured, multimodal, and hard to incorporate into current AI systems. This paper introduces the “k-agents” framework, designed to support experimentalists in organizing laboratory knowledge and automating experiments with agents. The framework uses large-language-model-based agents to encapsulate laboratory knowledge, including available operations and methods for analyzing results. To automate experiments, execution agents break multistep procedures into agent-based state machines, interact with other agents to execute steps, and analyze results. These results drive state transitions, enabling closed-loop feedback control. We demonstrate the system on a superconducting quantum processor, where agents autonomously planned and executed experiments for hours, successfully producing and characterizing entangled quantum states at human-level performance. Our knowledge-based agent system opens new possibilities for managing laboratory knowledge and accelerating scientific discovery.

Abstract:

Using quantum systems with more than two levels, or qudits, can scale the computational space of quantum processors more efficiently than using qubits, which may offer an easier physical implementation for larger Hilbert spaces. However, individual qudits may exhibit larger noise, and algorithms designed for qubits require to be recompiled to qudit algorithms for execution. In this work, we implemented a two-qubit emulator using a 4-level superconducting transmon qudit for variational quantum algorithm applications and analyzed its noise model. The major source of error for the variational algorithm was readout misclassification error and amplitude damping. To improve the accuracy of the results, we applied error-mitigation techniques to reduce the effects of the misclassification and qudit decay event. The final predicted energy value is within the range of chemical accuracy.

Abstract:

Blind quantum computation (BQC) protocols enable quantum algorithms to be executed on third-party quantum agents while keeping the data and algorithm confidential. The previous proposals for measurement-based BQC require preparing a highly entangled cluster state. In this paper, we show that such a requirement is not necessary. Our protocol only requires pre-shared Bell pairs between delegated quantum agents, and there is no requirement for any classical or quantum information exchange between agents during the execution. Our proposal requires fewer quantum resources than previous proposals by eliminating the need for a universal cluster state.

Abstract:

The volume and complexity of scientific literature are expanding rapidly, making it increasingly difficult to extract and synthesise information across studies. This challenge is particularly acute in the biological sciences, where evidence spans multiple levels of organisation and heterogeneous experimental designs. Large Language Model (LLM) pipelines offer a scalable route to evidence synthesis, but many existing approaches lack transparency, modularity, and effective mechanisms for human oversight. We present MetaBeeAI, an open-source, modular pipeline that integrates established LLM techniques into a coherent, auditable workflow for structured data extraction in biology. MetaBeeAI combines modular prompting, multi-pass extraction, and expert-in-the-loop validation within an interface that presents model outputs alongside source text, enabling inspection, correction, and iterative refinement. The pipeline produces machine-readable records of prompts, configurations, and expert annotations, supporting reproducibility and continuous improvement. We apply MetaBeeAI to 924 research papers on bees and pesticides, extracting structured information on species, compounds, exposure designs, and experimental context. Evaluation demonstrates improved consistency, convergence with expert judgement, and robustness across heterogeneous biological studies, highlighting the value of expert-guided refinement. MetaBeeAI provides a transparent and extensible framework for scalable evidence synthesis, supporting reliable integration of LLMs into biological research workflows.