Troubling trends in Interpretable Machine Learning Research - Amir Hossein Rahnama

Summit #15: Interpretable AI

In the post-deep learning era, since highly accurate models are now accessible and are running in production, we are faced with a new challenge: Will they perform as accurately on the new data they face out in the wild? Can we supervise them by "interpreting" their predictions? Can we "trust" their prediction? In this talk, I will present you some challenges of defining trust and interpretability in Machine Learning. Besides, I will present important works of research in the domain and how they approach interpretability and leave you with some open question to take home and reflect upon.

Presentation: https://drive.google.com/open?id=1m9MqlqUZmoRmG7vgL6SC6HB5bZC-dSbU

Тэги:

#interpretability #research #machine_learning

Скачать видео