The challenge of combining high-performance and energy efficiency is clearly present in today's electronics applications. This objective can be reached with Multiprocessor System-on-Chip (MPSoC) platforms that provide high-level of adaptability, performance, reliability and energy efficiency. Nevertheless, they have to be accompanied with a power management decision unit that continuously adapts their operation. This leads us to neural networks that can provide comparable properties to those of the cerebral cortex. We use an associative memory that relies on neural cliques to store information and sparse activity to retrieve it. We analyze these networks in power management applications using real-world data. We show that in order to be com...