What does an AI model “understand,” and why? Many look to easy-to-interpret neurons. E.g. neurons that activate for cat images are the “cat” part of the network. Surprisingly, we found that these class selective neurons can impair the performance of DNNs. https://t.co/rAF6y5aL2J pic.twitter.com/1N8RRi2mV4
— Facebook AI (@facebookai) October 28, 2020
from Twitter https://twitter.com/facebookai
October 28, 2020 at 09:13AM
via IFTTT
No comments:
Post a Comment