Goto

Collaborating Authors

 Deep Learning








Bridge the Modality and Capability Gaps in Vision-Language Model Selection

Neural Information Processing Systems

To better reuse the VLM resource and fully leverage its potential on different zero-shot image classification tasks, a promising strategy is selecting appropriate Pre-Trained VLMs from the VLM Zoo, relying solely on the text data of the target dataset without access to the dataset's images.



e0308d73972d8dd5e2dd27853106386e-Paper.pdf

Neural Information Processing Systems

Although deep learning programs havedemonstrated strong performance on novel applications, they sacrifice many of the functionalities of traditional software programs.