Sampled-data Synchronization of Recurrent Neural Networks with Multi-GPUs

This paper investigates the multi-rate sampled-data synchronization problem for recurrent reural networks with the multi-GPUs which have each different variable sampling rate. To handle the multi-GPU system with multi-sampling rate, the sampled-data sychronization error system is expressed as a summation of feedback subsystems with multi-sampling intervals. For the sampled-data controller design, Lyapunov functions with looped functions are constructed to use the information of the multi-rate sampling, and the modified free-matrix inequality is exploited to estimate the tighter upper bound intergral term. Finally, the simulation results show the effectiveness of the proposed method.