Recently, I came across an interesting non-linear visualization or dimensionality reduction method - tSNE(Distributed Stochastic Neighbour Embedding). Here is the original paper describing the method in much more detail. tSNE is just like PCA where you can reduce the dimensionality of data to 2 or 3 dimensions to visualize it. But unlike PCA, tSNE does that using non-linear ways. It converts everything to probability distributions. I am listing down the steps on a high level for a bigger picture. ... Read More

Note: All the code here is also available in this notebook. Variable names will be different in notebook then here. In this blog post we’ll talk about how different flavours of Gradient Descent help improve convergence of a function. Rosenbrock Function was taken for this experiment. It is a non-convex function, the one that we encounter regularly as Cost Functions in machine learning problems. Rosenbrock Function(Source: Wikipedia) Rosenbrock Function: $C(x, y) = (a-x)^2 + b(y-x^2)^2$ For this experiment values of both a and b is 1 Minimum value of Rosenbrock function then is 0 at (1, 1) Gradient of Rosenbrock Function w. ... Read More