Intelligent Adaptive Caching Middleware for Distributed Systems

Design a sophisticated caching middleware that implements adaptive caching strategies with machine learning-driven cache invalidation and predictive warming. Create a system that can dynamically adjust cache strategies based on access patterns, with support for multi-level caching (in-memory, distributed cache, persistent storage). Implement advanced eviction policies that consider access frequency, computational cost, and predicted future access patterns. Include comprehensive telemetry and self-tuning capabilities.