Categories

Algorithm

  • Dynamic 1D Piecewise Linear Spline Function Approx.

    The difficulty of understanding 1D ReLU-based Piecewise MLP guides us to work on Piecewise Linear Spline, its benefit for interpretation and ease of control. This experimental research starts with defining linear spline and deriving its gradient function. Finally, we create an algorithm to dynamically adjust the pieces of the linear spline to approximate some functions.

  • Perceptron to Deep-Neural-Network

    A Journey From Perceptron to Deep Neural Networks in a sequential fashion. Start with Perceptron, move to Logistic Regression, Single Layer Neural Network, Multilayer Perceptron (1 hidden layer) and finally to Deep Neural Network. Understand the algorithms sequentially along with visualization and math.

  • Artificial Neural Network Back Then

    Artificial Neural Network (ANN) is one of the most popular Machine Learning Algorithm. As the name suggests, the algorithm tries to mimic the Biological Neural Network, i.e. the Brain. In this post, we explore the development of the Algorithm from the very begining till development of Multilayer Perceptron.

  • Exploring Polynomial Regression

    Polynomial Regression is the generalization of Linear Regression. It is simple to understand but can do a lot. It is used to approximate any Non-Linear functions, which is almost always better than Linear Regression. Here, we extend the idea of curve fitting, learn its capacity, problems and its limitations.

  • Exploring Linear Regression

    Linear Regression is the simplest Machine Learning Algorithm and is fundamental to other algorithms such as Polynomial Regression and Neural Networks. Here, we explore the questions (What/Why/Where/How) of Linear Regression.


Research

  • Dynamic 1D Piecewise Linear Spline Function Approx.

    The difficulty of understanding 1D ReLU-based Piecewise MLP guides us to work on Piecewise Linear Spline, its benefit for interpretation and ease of control. This experimental research starts with defining linear spline and deriving its gradient function. Finally, we create an algorithm to dynamically adjust the pieces of the linear spline to approximate some functions.