Whether you’re detecting hot dogs or hotel reservations, deep learning is all the rage. When you pull back that pretty curtain made of pure math, though, guess what you find? That’s right: people. Lazy, unkempt, indecisive people, just waiting to make your beautiful functions useless. Join us as we explore the state of the art in machines not really learning what we think they’re learning, probably because of something we did.
This talk is more of a survey than a deep dive into a single paper, but a couple papers that go into detail on some methods we’ll explore toward the end of the talk (spoiler alert) are:
- Unmasking Clever Hans Predictors and Assessing What Machines Really Learn: https://arxiv.org/pdf/1902.10178.pdf
- AllenNLP Interpret: A Framework for Explaining Predictions of NLP Models: https://arxiv.org/pdf/1909.09251.pdf
Josh Ziegler fancies himself to be something other than a YAML engineer at Pylon. He spends his days eating peanuts and throwing the shells at ML researchers whose papers he’s reading.