Defending Quantum Classifiers against Adversarial Perturbations through Quantum Autoencoders — ThinkLLM