LLM-Guided ANN Index Optimization for Human-Object Interaction Retrieval
Researchers have developed a novel LLM-guided agent for optimizing Approximate Nearest Neighbor (ANN) index parameters in retrieval systems. This agent overcomes the limitations of traditional hyperparameter optimization methods by conditioning each proposal on the full optimization history, effectively navigating complex, coupled parameter spaces. Tested on the HICO-DET human-object interaction retrieval benchmark, the agent demonstrated significant performance gains, outperforming existing methods by over 33% and achieving a 15.3x throughput increase. AI
IMPACT This LLM-guided optimization method could significantly improve the efficiency and performance of various AI retrieval systems.