AURORA: Uncertainty-Aware Micro-Agent for Causal Observability in Computing Continuum
Researchers have introduced AURORA, a novel lightweight framework designed to diagnose and mitigate grey failures within edge-tier environments of the computing continuum. Grey failures often present ambiguous, overlapping symptoms that existing diagnostic tools struggle to address reliably due to limited causal awareness or high epistemic uncertainty, which can lead to destructive interventions. AURORA employs parallel micro-agents that integrate the free-energy principle, causal do-calculus, and localized causal state-graphs to perform counterfactual root-cause analysis within each fault's Markov blanket. By restricting inference to causally relevant variables, the system reduces computational overhead while maintaining diagnostic accuracy. A key feature is its dual-gated execution mechanism, which only authorizes local remediation when causal confidence is high and uncertainty is bounded; otherwise, it escalates the issue to the fog tier. Experimental results indicate that AURORA significantly outperforms baseline methods, achieving a 0% destructive action rate, 62.0% repair accuracy, and a mean time to repair of just 3 milliseconds, offering a robust solution for resilient edge computing systems.
Wire timeline
AURORA: Uncertainty-Aware Micro-Agent for Causal Observability in Computing Continuum
Researchers have introduced AURORA, a novel lightweight framework designed to diagnose and mitigate grey failures within edge-tier environments of the computing continuum. Grey failures often present ambiguous, overlapping symptoms that existing diagnostic tools struggle to address reliably due to limited causal awareness or high epistemic uncertainty, which can lead to destructive interventions. AURORA employs parallel micro-agents that integrate the free-energy principle, causal do-calculus, and localized causal state-graphs to perform counterfactual root-cause analysis within each fault's Markov blanket. By restricting inference to causally relevant variables, the system reduces computational overhead while maintaining diagnostic accuracy. A key feature is its dual-gated execution mechanism, which only authorizes local remediation when causal confidence is high and uncertainty is bounded; otherwise, it escalates the issue to the fog tier. Experimental results indicate that AURORA significantly outperforms baseline methods, achieving a 0% destructive action rate, 62.0% repair accuracy, and a mean time to repair of just 3 milliseconds, offering a robust solution for resilient edge computing systems.
cs.AI updates on arXiv.org