Automatic dynamic-system selection and configuration

Develop methods for automatically selecting the dynamic-system family and its configuration within dynamic-system-based neural-network model compression.

Background

The paper introduces a unified compression framework incorporating space-filling curves, chaotic systems, pseudo-random generators, and low-discrepancy sequences. The experiments evaluate manually selected configurations and demonstrate different compression–quality trade-offs across dynamic-system families. The authors explicitly identify automatic selection of the dynamic system and its configuration as an unresolved direction for future work.