Distributed intelligent systems rely on many autonomous agents that train a shared model on their own data, and a single faulty or malicious agent can corrupt that model. This paper proposes Quantum-Assisted Agentic Learning (QAAL), in which a coordinator agent perceives the agreement between agents' updates, keeps a trust memory of each agent, and plans which agents to aggregate by solving a quadratic unconstrained binary optimisation (QUBO) problem with a constraint-preserving quantum approximate optimisation algorithm (QAOA). QAAL was evaluated in simulation with 12 agents, non-IID data, random dropouts and three Byzantine attacks on two classification tasks, against FedAvg, coordinate-wise median, trimmed mean and Multi-Krum. With 4 of 12 agents faulty, QAAL reached 83.4% mean accuracy versus 55.2% for FedAvg and 72.5% for trimmed mean, and matched Multi-Krum (84.0%). The QAOA planner found the optimal agent subset in 98.7% of 1260 training rounds, and its trust memory separated honest from faulty agents with AUC up to 0.996. At the 12-agent scale tested, classical solvers are equally accurate and far faster, so no quantum speed-up is claimed.
Distributed intelligent systems rely on many autonomous agents that train a shared model on their own data, and a single faulty or malicious agent can corrupt that model. This paper proposes Quantum-Assisted Agentic Learning (QAAL), in which a coordinator agent perceives the agreement between agents' updates, keeps a trust memory of each agent, and plans which agents to aggregate by solving a quadratic unconstrained binary optimisation (QUBO) problem with a constraint-preserving quantum approximate optimisation algorithm (QAOA). QAAL was evaluated in simulation with 12 agents, non-IID data, random dropouts and three Byzantine attacks on two classification tasks, against FedAvg, coordinate-wise median, trimmed mean and Multi-Krum. With 4 of 12 agents faulty, QAAL reached 83.4% mean accuracy versus 55.2% for FedAvg and 72.5% for trimmed mean, and matched Multi-Krum (84.0%). The QAOA planner found the optimal agent subset in 98.7% of 1260 training rounds, and its trust memory separated honest from faulty agents with AUC up to 0.996. At the 12-agent scale tested, classical solvers are equally accurate and far faster, so no quantum speed-up is claimed.
Abstract The convergence of quantum optimization and autonomous agentic reasoning defines a new computational frontier in large-scale resource coordination. This work introduces Q-PACT, a Quantum-Parallel Agentic Coordination Toolkit that unifies quantum optimization, fairness-driven orchestration, and adaptive multi-agent intelligence into a reproducible hybrid framework. Q-PACT addresses a key limitation in distributed and cloud-based quantum infrastructures; inefficient, inequitable scheduling of scarce quantum resources among competing agents and tasks. The proposed architecture formulates multi-agent scheduling as a Quadratic Unconstrained Binary Optimization (QUBO) model, solved through a heterogeneous pipeline integrating quantum annealing (D-Wave), variational optimization (QAOA on IonQ/Azure), and classical meta-heuristics. A noise-aware encoder, solver-adapter layer, and agentic manager collaborate to maintain fairness, adaptability, and load balance, while the QML-AURA subsystem continuously refines solver hyperparameters through feedback-driven variational learning. Q-PACT’s hybrid orchestration layer enables dynamic re-optimization, fault-tolerant fallback execution, and secure artifact exchange across distributed backends, achieving resilience under the noise and queueing constraints of NISQ-era hardware. Empirical evaluations demonstrate significant improvements in makespan reduction, workload variance, and energy minimization compared with classical heuristics, establishing the viability of quantum-assisted agentic scheduling at scale. By bridging quantum optimization paradigms, multi-agent systems, and fairness-aware orchestration, Q-PACT provides a foundational blueprint for the next generation of autonomous, cloud-integrated, and quantum-enabled coordination systems.
Federated Learning (FL) trains a global model across decentralized clients while preserving data privacy, but at scale it is vulnerable to malicious updates. Byzantine-resilient aggregation methods such as MultiKrum score gradients against their nearest neighbors and can miss malicious updates that preserve the statistical properties of honest ones. We propose a quantum annealing approach that reformulates client selection as a Quadratic Unconstrained Binary Optimization (QUBO) problem, encoding pairwise distances into a cost function solved by quantum annealers (QA). Unlike MultiKrum's greedy per-client scoring, the QUBO formulation jointly optimizes over all subsets to find the mutually closest group of m clients. At small scale (15 clients), QUBO outperforms MultiKrum on the most challenging Byzantine attacks: e.g., Advanced LIE is detected with 95.11% accuracy versus 81.33% on MNIST and 97.78% versus 75.56% on CIFAR-10. QUBO fares poorly on simpler attacks where MultiKrum excels, so the two methods are complementary. QUBO quality also degrades as the number of clients grows. To address this, we introduce a MultiSignal ensemble that uses a dual-feature routing gate based on Euclidean and cosine Krum score gaps to classify attacks into four regimes and routes evasion attacks to a suspicion-penalized QUBO with agreement voting. At 100 clients on MNIST, MultiSignal achieves 95.3% average detection accuracy versus 91.8% for classical MultiKrum, with the largest gains on Sparse Lie (72.0% to 95.2%, +23.2 points) and Advanced Lie (80.4% to 85.2%, +4.8 points). These results show that QUBO-based quantum annealing with MultiSignal is a principled and scalable defense against the most challenging Byzantine strategies in federated learning.
Autonomous agents are being used in more complex and dynamic environments where efficient learning, adaptive decision-making, and scalable optimization are required. Although classical machine learning (ML) and reinforcement methods have greatly contributed to agent intelligence, they are usually restricted by constraints of computational complexity, scaling, and convergence to high-dimensional decision space. The current developments in quantum computing have created new possibilities to solve these problems using quantum ML (QML). In this chapter, the authors discuss the implementation of QML in collaboration with autonomous agents in the larger paradigm of agentic artificial intelligence (AI) and quantum AI agents. It offers a hybrid classical–quantum architectural model where quantum learning modules improve perception, policy learning, and decision optimization, and do not conflict with classical agent pipelines. The chapter analyzes application cases in autonomous robotics, intelligent transportation systems, multi-agent coordination, smart energy management, as well as communication networks, and the chapter shows the potential benefits of quantum-enhanced learning in these areas. The issue of quantum hardware neediness, data encoding, scaling, hybrid integration, and real-life application are critically evaluated and unanswered. In summary, the chapter is a well-organized approach to the purpose of QML in the development of the next-generation autonomous agent and an insight into creating practical, scalable, and intelligent quantum AI agents.
We propose a reinforcement learning (RL) scheme for feedback quantum control within the quantum approximate optimization algorithm (QAOA).We reformulate the QAOA variational minimization as a learning task, where an RL agent chooses the control parameters for the unitaries, given partial information on the system.Such an RL scheme finds a policy converging to the optimal adiabatic solution of the quantum Ising chain that can also be successfully transferred between systems with different sizes, even in the presence of disorder.This allows for immediate experimental verification of our proposal on more complicated models: the RL agent is trained on a small control system, simulated on classical hardware, and then tested on a larger physical sample.
Reinforcement Learning–based Quantum Approximate Optimization Algorithm (QAOA-RL) is used to solve the Vehicle Routing Problem (VRP) by transforming it into a Quadratic Unconstrained Binary Optimization (QUBO) form. The RL agent learns to optimize QAOA parameters for better route selection. Results show that QAOA-RL achieves a lower total cost compared to exact ILP and heuristic Greedy Search methods. Although the training time is higher, the solution quality improves significantly for larger problem sizes. This study highlights QAOA-RL as a scalable and promising hybrid quantum–classical approach for real-world logistics optimization.
Optimization models for large smart grids need to consider network constraints, renewable energy uncertainty, storage dynamics and flexible demand simultaneously. The mixed discrete-continuous nature of these problems is not suitable for the operational time constraints, and a single quantum processor cannot yet handle real grid instances. A distributed quantum computing framework for an electrical network is proposed in this paper, which divides the network into coupled regions, assigns compact quadratic unconstrained binary optimisation subproblems to heterogeneous quantum workers, and employs an asynchronous classical coordinator to enforce boundary consensus. The four components of the framework are adaptive penalty encoding, warm-started variational circuits, confidence-aware result aggregation and projection-based feasibility repair. A reproducible simulation study is constructed for modified 33-bus, 118-bus, 300-bus and 1354-bus systems with renewable generation, batteries and responsive loads. In all the test cases, the proposed method reduced the operating cost and constraint violation compared with centralised QAOA, quantum annealing, ADMM, and mixed-integer programming under the same time budget. It still performs reasonably well under high worker latency and depolarising noise. Based on the above results, near-term quantum resources will be more credible as coordinated regional accelerators than as a single grid optimizer. Therefore, it will be presented as an engineering architecture, an explicit optimisation map and a multi-metric evaluation system for scalable quantum-assisted smart grid operation.