Researchers have developed a machine learning approach to compressing image data with greater accuracy than learning-free computation methods, with applications for retinal implants and other sensory prostheses.