Ensemble Neural Representation Networks

Abstract Implicit Neural Representation (INR) has recently attracted considerable attention for continuous characterization of various types of signals. The existing INR techniques require lengthy training processes and high-performance computing. In this letter, we propose a novel ensemble architecture for INR that resolves the aforementioned problems. In this architecture, the representation task is divided into several sub-tasks done by independent sub-networks. We show that the performance of the proposed ensemble INR architecture may decrease if the dimensions of sub-networks increase. Consequently, it is vital to suggest an optimization algorithm to find suitable structures for the ensemble networks, which is also done in this paper. According to the simulation results, the proposed architecture not only has significantly fewer floating-point operations (FLOPs) and less training time, but it also has better performance in terms of Peak Signal to Noise Ratio (PSNR) compared to those of its counterparts. (The source code is available at https://github.com/AlirezaMorsali/ENRP)