BROWSE
Detail
Conference Paper
FORE-Twin: Neuro-Symbolic Middleware for Latency-Aware Data Staging in Federated Digital Twin Systems
Cited - time in 
Share 
Share
- Authors
- Jiwoo Han, Yangkoo Lee, Kyounghyun Park, Daesub Yoon
- Issue Date
- 2026-05
- Citation
- International Symposium on Cluster, Cloud and Internet Computing (CCGrid) 2026, pp.37-40
- Publisher
- IEEE
- Language
- English
- Type
- Conference Paper
- Abstract
- Federated digital twin (DT) ecosystems require data sharing across administrative boundaries. Due to strict data governance and heterogeneous schemas, bulk data replication is often impractical, making query-driven, synchronous cross- silo access the dominant pattern. This creates severe WAN latency bottlenecks. We present FORE-Twin, a neuro-symbolic middleware for proactive, latency-aware data staging in federated DTs. To bound overhead, FORE-Twin models runtime execution as a dynamic, multi-level dependency graph. It employs a hybrid engine: a deterministic Rule-Based Engine (RBE) enforces structural and governance constraints for baseline stability, while a Hierarchical Reinforcement Learning (HRL) agent learns complex cross-domain access patterns to optimize prefetching. Evaluated on a WAN-emulated testbed with real-world smart parking traces, FORE-Twin improves inter-system cache hit rates to 65.7% and reduces average access latency by 34% compared to the strongest learning baseline. By synergizing symbolic safety with neural adaptivity, FORE-Twin mitigates RL cold-start instability and delivers practical WAN latency masking
- Keyword
- Federated Digital Twins, Neuro-Symbolic Middleware, Distributed Data Staging, Hierarchical Reinforcement Learning
- KSP Keywords
- Access Latency, Access pattern, Cold-start, Cross Domain, Data Replication, Digital Twin, Hierarchical reinforcement learning, Hybrid engine, Multi-level, Query-driven, Real-world