LLM-Guided Initialization Speeds Quantum Medical Image Classification

Using LLMs to propose initial parameters for quantum neural networks (AdaInit) achieves 160x faster convergence compared to random initialization. On the DMR-IR mammography dataset, the method maintains 61.4% accuracy while significantly increasing gradient variance.