We present an advanced approach to medical question-answering (QA) services, using fine-tuned Large Language Models (LLMs) to improve the accuracy and reliability of healthcare information. Our study focuses on optimizing models like LLaMA-2 and Mistral, which have shown great promise in delivering precise, reliable medical answers. By leveraging comprehensive datasets, we applied fine-tuning techniques such as rsDoRA+ and ReRAG. rsDoRA+ enhances model performance through a combination of decomposed model weights, varied learning rates for low-rank matrices, and rank stabilization, leading to improved efficiency. ReRAG, which integrates retrieval on demand and question rewriting, further refines the accuracy of the responses. This approach enables healthcare providers to access fast, dependable information, aiding in more efficient decision-making and fostering greater patient trust. Our work highlights the potential of fine-tuned LLMs to significantly improve the quality and accessibility of medical information services, ultimately contributing to better healthcare outcomes for all.

Smart objects in the Social Internet of Things (SIoT), such as cameras, drones, and vehicles, exchange images that act as visual evidence and drive automated decisions. These images can be altered in their pixels or their metadata, replayed, or injected by unauthorized publishers. Central verification services can check them, but such services must be reachable at verification time, form a bottleneck, and observe who produced which image. This paper presents a decentralized scheme that makes an SIoT image object self-verifiable, so that an intermittently connected verifier can check it offline. Each object binds its image hash, its name, its provenance record, and its source identity in one signature, and it carries its own key documents and authorization chain. Trust anchors are Decentralized Identifiers computed as key thumbprints, so no registry, ledger, or certificate authority is queried during verification. Our central technical point is that a signed hash alone gives only name-bound replay prevention. We therefore add temporal layers that such schemes usually omit: a freshness mechanism with interactive, beacon, and transparency log variants; signed status lists with a proven bounded staleness revocation guarantee; and monotone epochs that resist rollback of rotated keys and documents. We prove the base goals by reduction to signature unforgeability and hash collision resistance under a Dolev–Yao adversary, and we prove the temporal properties as unbounded inductive invariants discharged in Z3. An Ed25519 and SHA-256 implementation verifies a typical image in under two milliseconds with about 1.6 kB of metadata. The evidentiary levels are stated separately and are not interchangeable. The base object is proven, implemented, and measured; the freshness, revocation, and rollback layers are proven but not measured; the pseudonymous mode is design-only. “Self-verifiable” means that provenance is checked cryptographically from the object and one anchor. It does not mean the scheme proves that the captured scene is real, and it is conditional on a preconfigured root identifier and, for the temporal layers, on a status list or time source.

The complexities of healthcare data, including privacy concerns, imbalanced datasets, and interoperability issues, necessitate innovative machine learning solutions. Swarm Learning (SL), a decentralized alternative to Federated Learning, offers privacy-preserving distributed training, but its reliance on blockchain technology hinders accessibility and scalability. This paper introduces a \textit{Simplified Peer-to-Peer Swarm Learning (P2P-SL) Framework} tailored for resource-constrained environments. By eliminating blockchain dependencies and adopting lightweight peer-to-peer communication, the proposed framework ensures robust model synchronization while maintaining data privacy. Applied to cancer histopathology, the framework integrates optimized pre-trained models, such as TorchXRayVision, enhanced with DenseNet decoders, to improve diagnostic accuracy. Extensive experiments demonstrate the framework's efficacy in handling imbalanced and biased datasets, achieving comparable performance to centralized models while preserving privacy. This study paves the way for democratizing advanced machine learning in healthcare, offering a scalable, accessible, and efficient solution for privacy-sensitive diagnostic applications.