This small animation was made with Chirag Gupta (IIT Kanpur) depicting the training of the perceptron algorithm for a random dataset. The right plot depicts the points which the algorithm actually uses for training, thus illustrating that the effective number of points required are much less than the actual dataset. This also shows that more points are picked near the margin and thus motivates for margin based active learning.
Video Link : https://www.youtube.com/watch?v=xpJHhHwR4DQ