Friday, April 20 at 3:00 PM
We consider a central problem from pattern recognition. Given a set of examples of objects belonging to various classes, decide the class to which a new object belongs. The set of objects is called a “”training set”” and is represented as a set of points in the Euclidean space, and their class is represnted by a colour assigned to each point. The new object is our query point, which we want to colour.
A popular method for colouring query points is the nearest neighbour rule, i.e. the new point gets the colour of the point of the training set nearest to it. However, in a practical scenario, the training set is often too large and hence operations on it become costly in terms of time and storage. So, we try to preprocess the training set and reduce its size. In this talk we discuss some heuristics and results for the same, such that the probability of assigning a wrong colour to a query point is low.
Download presentation (password protected)