Communication is the dominant source of energy consumption in Internet-of-Things (IoT) networks, yet many sensed measurements exhibit strong temporal correlations and provide little new information to the receiver. This paper introduces \textsc{ADAPTIVEML}, a volatility-aware predictive communication framework that enables IoT devices to intelligently decide when communication is necessary. Each sensor maintains a lightweight machine learning predictor and transmits only when the prediction residual exceeds an adaptive threshold proportional to the local signal volatility. By normalizing prediction errors using a rolling estimate of signal variability, the proposed transmission policy automatically adapts to changing environmental conditions, seasonal variations, and deployment-specific dynamics without manual threshold tuning. To address long-term non-stationarity, we further propose \textsc{ADAPTIVEML-RLS}, an online learning extension based on Recursive Least Squares (RLS) with exponential forgetting, allowing continuous adaptation to sensor drift and evolving signal characteristics. Extensive experiments are conducted on three heterogeneous real-world datasets comprising more than 2.4 million sensor observations from outdoor environmental monitoring, indoor wireless sensor networks, and urban air-quality sensing. Compared with six representative baselines, including periodic transmission, static-threshold suppression, ARIMA, Kalman filtering, EMA, and LMS filtering, \textsc{ADAPTIVEML} achieves up to 94.7\% transmission reduction while maintaining a reconstruction error of 0.352$^\circ$C. \textsc{ADAPTIVEML-RLS} further reduces reconstruction error by 12--18\% under drift conditions while preserving transmission reduction above 93\%. These results demonstrate the effectiveness of volatility-aware predictive communication for energy-efficient and adaptive IoT networks.
J. Kangethe, Ifrat Ikhtear Uddin, Longwei Wang· 0 citations
Large language models (LLMs) exhibit remarkable capabilities but remain highly vulnerable to adversarial prompts and jailbreak attacks. Existing approaches primarily analyze these failures through input-output behaviors or attribution methods, offering limited insight into how adversarial perturbations alter the model's internal reasoning. Consequently, the mechanisms underlying unsafe or incorrect behaviors remain poorly understood. We introduce a mechanistic framework for diagnosing LLM vulnerabilities using paired internal computation graphs, which represent prompt-specific inference as structured causal interactions among latent features. By constructing and aligning computation graphs for clean and attacked prompts, we reveal that adversarial attacks induce systematic transformations of internal reasoning, including suppression of safety-relevant components, emergence of attack-specific features, and rerouting of computation paths. Building on this representation, we propose a unified framework that (i) decomposes computation into invariant, suppressed, and emergent structures, (ii) identifies recurring vulnerability motifs associated with failure modes, and (iii) performs causal interventions on nodes, paths, and subgraphs to directly evaluate their contributions to attack success. This enables a transition from descriptive attribution to causal diagnosis of model failures. Experiments across multiple open-source LLMs and diverse adversarial and jailbreak benchmarks demonstrate that structural deviations in internal computation graphs strongly correlate with unsafe behaviors. Furthermore, targeted interventions on identified vulnerability motifs improve model robustness, establishing internal computation graphs as a principled foundation for understanding, diagnosing, and mitigating LLM vulnerabilities.