Artificial Intelligence & Machine Learning

Machine Learning is a concept which deals with how to train a computer to ‘discover and learn’ patterns in a data set without being explicitly programmed for it. Over the last decade, there has been a tremendous proliferation of datasets which provide raw data for correlation analysis. A typical Machine Learning framework is provided with a large amount of data and is tasked with discovering patterns within it. To help it, one might choose to label the ‘input’ data. For example, we might give it 10,000 images of faces and label which ones are men and which ones are women. We call this ‘supervised learning’. The system then proceeds to figure out patterns and relationships that converge to making sure it is accurately able to relate these patterns to the labels provided. In this approach, the system is provided with a ‘correct dataset’ to learn in the hope that the learning from this ‘training set’ can be applied to ‘unknown but similar’ data sets in the future. Another example is how Bayesian filters work on email spam – we users keep flagging emails as spam which it uses as a learning tool (training data) to be able to mark future emails as spam.

Another form is to provide the system with a large amount of data and asking it to figure our relationships without guidance. This is called ‘unsupervised learning’ An example of this could be ‘here is 5 years worth of data for what a section of users search for on the internet. 5 years later these users had cancer. Use this data to find out if we can identify key common early symptoms of cancer, if possible’. (This was not a random example, Microsoft Research did something similar)