Data Neutronium

Just as solid neutronium is the densest form of ordinary matter, data neutronium is the most highly compressed kind of data -- it is pure information. This project is developing a system to compress data such as imagery and speech into data neutronium. This project is developing a system to compress data such as imagery and speech into data neutronium. We are using two new results: a theory relating the tolerable level of reconstruction error to the maximum possible level of compression, as the Cottrell/Munro/Zipser neural network technique (which is capable of achieving maximum compression). Unlike fractal compression (the only other known method theoretically capable of approaching data neutronium density levels), which is computationally impractical, the C/M/Z neural network (which is trained off-line once) is computationally simple, and can carry out both data compression and decompression in real-time using low-cost hardware. Our goal is to achieve compression ratios from 100:1 to 1000:1 for speech and imagery with broadcast quality reconstruction.