Zero Shot Classification Task

Zero-shot classification aims to enable models to classify data points into categories never seen during training, leveraging existing knowledge to recognize novel concepts. Current research focuses on improving the adaptability of vision-language models like CLIP, often through test-time adaptation techniques or by incorporating external knowledge sources like textual descriptions or retrieved data. These advancements are significant because they reduce the reliance on extensive labeled datasets, making machine learning more efficient and applicable to domains with limited annotated data, impacting fields like image recognition, information extraction, and assistive technologies.