Towards trustworthy and personalized situation-aware services in IoT-enabled and learning-based smart systems

As we steadily advance into the era of situation-aware (situ-aware) ubiquitous computing, IoT-enabled smart environments continue to expand, and increasingly rely on learning-driven pipelines to synthesize situations from distributed sensing contexts and to support adaptive, service-oriented actuation. Despite rapid progress, two core obstacles limit the realization of trustworthy and personalized situation-aware services: (i) data scarcity and heterogeneity across edge domains, which weakens centralized Federated Learning (FL) under non-i.i.d. distributions; and (ii) the lack of verifiable, tamper-resistant coordination in vendor-specific IoT stacks, where cloud-centric control introduces integrity and accountability risks. To address these challenges, this dissertation presents an approach towards trustworthy and personalized situ-aware services in IoT-enabled, learning-based smart systems. First, we propose Situ-Oracle, a blockchain–oracle framework that provides situ-analysis as a service for local BIoT. In Situ-Oracle, smart contracts serve as deterministic, auditable coordination and access-control channels, while an external computation oracle enables scalable learning inference/training; we further instantiate the service using recurrent neural network (RNN)-based models. Second, we study a decentralized, clustering-based mutual-learning scheme that enables iterative local model refinement via intra-cluster knowledge sharing for distribution-aware personalization. Finally, we extend mutual learning towards adversarial settings by introducing a verifiable blockchain-based framework leveraging local endorsement, a hybrid blockchain architecture, and zero-knowledge proofs to validate shared knowledge without disclosing raw data. Evaluations demonstrate improved situ-analysis accuracy under heterogeneity, reduced communication overhead, and improved integrity guarantees compared to baseline FL and existing blockchain-based learning approaches.

Committee: Wensheng Zhang (co-major professor), Carl Chang (co-major professor), Ying Cai, Simanta Mitra, and Chenglin Miao