At Jane Street, we often work with data that has a very low signal-to-noise ratio, but fortunately we also have a lot of data. Where practitioners in many fields might be accustomed to having tens or hundreds of thousands of correctly labeled examples, some of our problems are more like having a billion training examples
