Abstract: Existing learning-based methods to compress PCs attributes typically employ variational autoencoders (VAE) to learn compact signal representations. However, these schemes suffer from limited ...