One Smooth Stone: How to Beat Goliath with Quality Training Data

As Data Scientists, we have become accustomed to thinking that a better model accuracy implies injecting more data. Deep Learning models, for example, are traditionally known to be extremely data-greedy, but it seems that collecting a lot of data is the necessary price to pay to obtain a high-performance model. However, in the age of Big Data, this doesn’t seem like a major issue. The caveat though, is that Supervised Machine Learning algorithms require structured, labeled data, and the process of labeling data is unfortunately neither an easy nor a cheap task. Thankfully, there is a better way: Active Learning. In this talk, I will dig into the concepts of Semi-Supervised Learning and Active Learning before explaining how Figure Eight helps companies make AI work in the real world at a fraction of the labeling cost.