Probability Quality

Probability quality research explores the relationship between the probability assigned to an output by a model (e.g., a generated image, text, or manufactured product) and its perceived quality by human observers. Current research focuses on developing methods to improve this relationship, including using deep learning architectures like Koopman models and variational autoencoders to better capture complex quality dynamics and uncertainty, and employing novel evaluation metrics that go beyond simple likelihood scores. This work is crucial for advancing various fields, from manufacturing quality control to the development of more human-like AI systems, by enabling the creation of more reliable and high-quality outputs.