Skepticism in Deep Learning
Explore the relationship between accuracy and confidence in deep learning classifiers, and the impact of training corpus on potential weaknesses. Anh explains that accuracy and confidence are not equivalent, and that the discriminated nature of the network is the root of the problem. The discussion also touches on the limitations of deep learning in providing human-readable explanations for classification decisions.In this clip
From this podcast
Data Skeptic
Easily Fooling Deep Neural Networks
Related Questions