<ul><li>The researchers propose a novel approach called hyper-compression for model compression.</li><li>Hyper-compression represents the parameters of the target network using dynamic systems as hyperfunctions.</li><li>This approach offers a preferable compression ratio, no post-hoc retraining, affordable inference time, and short compression time.</li><li>The hyper-compression method achieves close-to-int4-quantization performance with less than 1% performance drop.</li></ul>

Hyper-Compression: Model Compression via Hyperfunction

Discover more