Charla "Uncertainty in Machine Learning - Large Vision-Language Models and Computer Visión"

What if we train a model to classify dogs and cats, but it is later tested with an image of a human? Generally the model will output either dog or cat, and has no ability to signal that the image has no class that it can recognize.

Machine learning models by default do not provide estimates of their confidence or uncertainty, which hinders their use in applications involving humans. Possible solutions is the use of Bayesian Neural Networks or similar models.

In this talk I will show research applications of neural networks with uncertainty quantification, covering Computer Vision, Large Language Models and Vision-Language Models. This includes super-resolution, frame generation, verbalized uncertainty, robustness to corrupted inputs, and input uncertainty.