UCR researchers develop a method called Test-Time Matching, an approach that significantly improves how AI systems interpret relationships between text and images, especially when presented with unfamiliar combinations.