Compression and decompression for neural networks

A data processing system 200 for implementing convolutional processes includes: a convolution engine 202 and a set of weight decoders 212 including a first weight decoder and a second weight decoder that implement a first decompression function and a second decompression function, respectively. A weight decoder selection module 216 for selecting a weight decoder from the set of weight decoders is provided. The data processing system, receives a compressed set of weight values 218, selects a weight decoder using the weight decoder selection module, and processes the compressed set of weight values using the selected weight decoder to obtain an uncompressed set of weight values 204. The uncompressed set of weight values are provided to the convolution engine. A corresponding set of weight encoders including a first and second encoder using a first and second compression function is also disclosed. Weight decoders are associated to characteristics including compression ratio and measure of computational resources. Decompression functions include using a look-up table.