Back
Cees Snoek (University of Amsterdam)
Author/s
About the resource/s
In the past decade, artificial intelligence has made remarkable progress, achieving feats like self-driving cars, defeating go-masters, and precise image categorisation through supervised deep learning with labelled data. However, a significant challenge is that deep learning models tend to be biased towards their training conditions and struggle in real-world situations that differ from their training data in terms of data, labels, and objectives. Simply increasing the scale of training is not a viable solution due to resource constraints and human ability to generalise efficiently with less data.
Documents
Other Sources
