Researchers Open-Source WiSE-FT Algorithm For Fine Tuning AI Models

When making zero-shot inference, large pre-trained models like CLIP or ALIGN provide consistent accuracy across various data distributions (i.e., without fine-tuning on a specific dataset). While existing fine-tuning methods vastly improve accuracy on a given target distribution, they frequently compromise robustness to distribution shifts. This conflict can be resolved by presenting a simple and effective strategy for enhancing robustness while fine-tuning: assembling the zero-shot and fine-tuned models (WiSE-FT). 

An approach for fine-tuning AI models that enhance robustness during distribution shift …

ai ai models algorithm artificial researchers