SOLAR Framework Optimizes Semantic Retrieval Buffers for LLM Agents

The SOLAR framework replaces ineffective classic cache policies like LRU and LFU with a learning-augmented approach for semantic workloads. By using Bayesian online learning and regret accumulation, it addresses the lack of temporal locality in embedding-based retrieval buffers.