Quantum machine learning (QML) has attracted growing interest for their ability to achieve superior performance with significantly fewer parameters. However, the high cost and scarcity of current hardware push inference to cloud-hosted quantum devices, creating a tension between verifiability and confidentiality. This work proposes a novel framework that converts quantum neural network operations into classical arithmetic circuits that faithfully approximate genuine quantum computations. By encrypting these circuits with zero-knowledge proofs, it ensures computational validity while concealing internal parameters. Experimental results show that our classical circuits achieve fidelity above 0.9996 and total variation distance below 1% compared to actual quantum computations, verifying the practicality of trustworthy and privacy-preserving quantum inference.

Federated learning (FL) has emerged as a powerful paradigm for decentralized training, particularly in privacy-sensitive fields such as medical Internet of Things (IoT) services, where data security is important. However, FL faces challenges due to nonindependent and identically distributed (non-IID) data among clients, which can lead to suboptimal performance. In addition, there is a risk of data leakage during the aggregation process. To address these issues, we propose a novel approach using an auction mechanism to filter out unreliable clients, ensuring that only trustworthy participants are involved in the learning process. The selected clients are organized in a ring topology, eliminating the need for a central server and thereby reducing the risk of data breaches. Additionally, we leverage quantum neural networks (QNNs) to enhance security further, utilizing the quantum no-cloning theorem to prevent the duplication of quantum parameters. The results demonstrate that our approach can handle non-IID data distributions effectively and improve model performance, even with small and imbalanced datasets.

Quantum federated learning (QFL) incorporates the principles of quantum neural networks (QNNs) and federated learning (FL), ensuring privacy by updating quantum parameters to the server. However, server aggregation in general QFL involves quantum parameters from all local models, leading to performance degradation due to the participation of outlier heterogeneous local models. Inspired by this, a novel QFL algorithm is proposed to enhance learning performance by excluding outliers before aggregation using Lyapunov optimization at QFL server. This Lyapunov optimization is used to take care of the tradeoff between accuracy and latency, where the accuracy is associated with the freshness of aggregation. Additionally, an entropy-and-fidelity aware algorithm is proposed, which relies on the degree control to select outliers for exclusion comprehensively. This algorithm addresses data and quantum aspects by evaluating the imbalance of data classes in each local model using entropy and quantifying the dissimilarity in quantum states between the target model and each local model through fidelity. Experimental results demonstrate that the proposed algorithm outperforms benchmarks, effectively excluding heterogeneous local models to improve performance while ensuring stability.