ETD

• application placement
• cloud-edge computing
• deep reinforcement learning
• graph neural networks
• neuro-symbolic AI
• symbolic reasoning

This work investigates the problem of Cloud-Edge application placement, namely the allocation of distributed application components across a heterogeneous computing continuum that ranges from centralized Cloud datacenters to resource-constrained Edge devices. The problem is intrinsically complex due to its NP-hard nature and the need to satisfy strict resource and network constraints while optimizing multiple quality metrics such as latency, bandwidth utilization, and cost efficiency. Symbolic approaches, exemplified by FogBrainX, ensure constraint compliance but exhibit prohibitive computational costs on large-scale infrastructures. Conversely, neural approaches, such as FlagVNE, enable rapid inference and good scalability but often fail in highly constrained or saturated scenarios. To address these limitations, this work introduces FogBlend, a neuro-symbolic methodology that integrates neural inference with symbolic reasoning. Specifically, FogBlend leverages FlagVNE to provide initial placements and invokes FogBrainX to correct invalid allocations through continuous reasoning. Both methods have been extended and combined within a dedicated Python-based framework. Experimental evaluations on realistic benchmark scenarios show that FogBlend achieves a favorable trade-off among inference time, placement quality, and success rate, thereby demonstrating its effectiveness and scalability compared to purely symbolic or neural solutions.