#denoising-autoencoders News & Analysis

Noise-Aware Visual Representation Learning for Medical Visual Question Answering

Researchers propose a noise-aware medical visual question answering framework that uses denoising autoencoders to improve the robustness of visual representations when connecting vision encoders to large language models. The approach achieves competitive performance on medical imaging benchmarks while demonstrating enhanced resilience to noisy inputs through parameter-efficient fine-tuning.