Starts in


February 15, 2022

11:00 am / 12:00 pm

ABSTRACT: Let us consider a difficult computer vision challenge. Would you want an algorithm to determine whether you should get a biopsy, based on an x-ray? That’s usually a decision made by a radiologist, based on years of training. We know that algorithms haven’t worked perfectly for a multitude of other computer vision applications, and biopsy decisions are harder than just about any other application of computer vision that we typically consider. The interesting question is whether it is possible that an algorithm could be a true partner to a physician, rather thanmaking the decision on its own. To do this, at the very least, we wouldneed an interpretable neural network that is as accurate as its black boxcounterparts. In this talk, I will discuss two approaches to interpretable neural networks: (1) case-based reasoning, where parts of images are compared to other parts of prototypical images for each class, and (2) neural disentanglement, using a technique called concept whitening. The case-based reasoning technique is strictly better than saliency maps, and theconcept whitening technique provides a strict advantage over the posthoc use of concept vectors. Here are the papers I will discuss:

This Looks Like That: Deep Learning for Interpretable Image Recognition. NeurIPS spotlight, 2019. https://arxiv.org/abs/1806.10574
IAIA-BL: A Case-based Interpretable Deep Learning Model for Classification of Mass Lesions in Digital Mammography, 2021. https://arxiv.org/abs/2103.12308
Concept Whitening for Interpretable Image Recognition. Nature Machine Intelligence, 2020. https://rdcu.be/cbOKj
Stop Explaining Black Box Machine Learning Models for High Stakes Decisions and use Interpretable Models Instead, Nature Machine Intelligence, 2019. https://rdcu.be/bBCPd
Interpretable Machine Learning: Fundamental Principles and 10 Grand Challenges, 2021 https://arxiv.org/abs/2103.11251

BIO: Coming Soon