Web: http://arxiv.org/abs/2201.11194

Jan. 28, 2022, 2:10 a.m. | Arjun R Akula, Song-Chun Zhu

cs.LG updates on arXiv.org arxiv.org

Attention based explanations (viz. saliency maps), by providing
interpretability to black box models such as deep neural networks, are assumed
to improve human trust and reliance in the underlying models. Recently, it has
been shown that attention weights are frequently uncorrelated with
gradient-based measures of feature importance. Motivated by this, we ask a
follow-up question: "Assuming that we only consider the tasks where attention
weights correlate well with feature importance, how effective are these
attention based explanations in increasing human …

arxiv attention

More from arxiv.org / cs.LG updates on arXiv.org

Director, Data Science (Advocacy & Nonprofit)

@ Civis Analytics | Remote

Data Engineer

@ Rappi | [CO] Bogotá

Data Scientist V, Marketplaces Personalization (Remote)

@ ID.me | United States (U.S.)

Product OPs Data Analyst (Flex/Remote)

@ Scaleway | Paris

Big Data Engineer

@ Risk Focus | Riga, Riga, Latvia

Internship Program: Machine Learning Backend

@ Nextail | Remote job