The most important challenge that researchers in machine learning face is that of the data. How do you decide whether the data that you have is good enough? When do you say the amount of data you have is enough? These are some questions that are very difficult to answer. The main purpose of building machine learning systems is to make them as general as possible for a given scenario. Say we are trying to build a handwritten digit classification system. Can we be sure that any image of digit 5 that we provide to the system will be able to output the correct output? Suppose our training data has hand samples from five people and we test the system on digits written by a sixth person. Will it perform as good? An extension to this problem is the lack of labeled data. One of the most laborious parts of collecting data is labeling data. Over time, with collaborative efforts from researchers, there are some good quality datasets available. But if you want to apply...