Neural networks use immense datasets to train against. While these are very sophisticated, they also prove to be brittle against outliers. A recent piece by researcher Janelle Shane shows just how problematic the presence (or lack thereof) of humble sheep can prove.
It’s become common now for IT companies to list deep learning algorithms as a major platform feature, from analytics to automation. But home does deep learning compare to actual human intelligence? Ray Lucchesi looked at some of its issues in the context of the MIT Intelligence Quest.