<p>This paper infers a novel quadratic inequality and formulates a new type of system titled fractional-order lower-dimension-valued bilayer neural networks (FOLDVBNNs). On one hand, based on the participation of both variable terms and the conjugates of variable terms, a new amplified inequality is set up in order to effectively combine the advantages of the non-separation method and the direct real-valued decomposition approach. On the other hand, the establishment of FOLDVBNNs can be carried out from a completely new and standardized perspective because the unified lower-dimension-valued system contains both real-valued bilayer neural networks and complex-valued bilayer one in fractional-order field. Generally, the newly established inequality is applied to address the completely new and relaxed criteria for the Mittag-Leffler synchronization issue of the new system of FOLDVBNNs, respectively. Further, this paper uses the particle swarm algorithm (PSO) to obtain the optimal solution of the multiple criteria. Finally, three numerical simulations are listed to provide the availability and progress of the acquired results and show the realization of the synchronous recovery of noisy complex signals through the model of FOLDVBNNs.</p>