Compression of UV spectrum with recurrent neural network
Leong Kwan Li, Ka Fai Cedric Yiu · 2010
In order to save time or storage space, compression techniques are applied. Recently compression techniques based on approximation theory are dominated by the fast Fourier and the wavelet transforms if noise is tolerated. For a given sequence, the compressed signal is represented as a linear sum of basic functions. In this note, we introduce a dynamical system approach for signal compressions. We demonstrate how to compress a UV spectrum by a discrete-time recurrent neural network. As an initial valued problem, the parameters we stored are the connection weights of the neural network and also the initial states. Compression ratio is also discussed. Storage space and energy is saved if good compression techniques are applied.