Zero Shot Classification

Zero-shot classification aims to classify data into categories unseen during model training, leveraging pre-trained models and prompting techniques to achieve this. Current research focuses on improving accuracy and robustness across diverse data types (images, text, audio, point clouds) using large language models (LLMs), vision-language models (VLMs) like CLIP, and various adaptation strategies such as prompt engineering, model label learning, and data augmentation. This field is significant because it reduces the reliance on extensive labeled datasets, enabling efficient classification in low-resource settings and facilitating applications in areas like medical image analysis, social media monitoring, and astronomical image processing.