CSE-574: Intro Machine Learning
Problem: The project is about different classification and regression techniques. There are six sub-problems under this project.
- Experiment with
Gaussian Discriminatorsby implementing Linear Discriminant Analysis (LDA) and Quadratic Discriminant Analysis (QDA) - Linear Regression implementation
- Experiment with Ridge Regression
- Using Gradient Descent for Ridge Regression
- Non-linear Regression
- Compare and interpret the results of different classification and regression methods
Problem: Handwritten Digits Classification using Multilayer Perceptron (MLP) Neural Network
Approach:
- Implemented MLP Neural Network from scratch to classify handwritten digits from MNIST dataset (achieved test accuracy of 93.45%)
- Used Feed Forward and Back Propagation to implement Neural Network
- Experimented the effect of
regularizationin thebias-variancetrade-off - Used the same neural network for more challenging face dataset and compared it with
Deep Neural NetworkandConvolutional Neural Networkusing the Tensorflow library
Problem: The project is about Support Vector Machine (SVM) and mulit-class logistic regression.
Approach:
- Implemented Logistic Regression (using one-vs-all strategy) to classify handwritten digit images into correct corresponding labels. In particular, built 10 binary-classifiers (one for each class) to distinguish a given class from all other classes.
- Developed multiclass logistic regression classifier and SVM for handwritten digits classification on the MNIST Dataset
