Learning with Prompts
Kate explains the concept of token embeddings and how prompts can be learned generically for various classes, enhancing zero-shot learning capabilities. She discusses the challenges of evaluating their model against existing ones, particularly those that don't leverage large vision language models like CLIP. Despite the complexities of data fairness and transparency, the superior performance of these models is undeniable.In this clip
From this podcast
The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence)
More Language, Less Labeling with Kate Saenko - #580
Related Questions