Projects were done in groups of three.
Compared the performance of K-Nearest Neighbors and Decision Tree across two benchmark datasets; Examined the impact of different cost functions, distance functions, and stopping criteria on accuracy; Experimented with crossvalidation, hyperparameter tuning, and different methods of feature selection
Investigated the performance of the logistic regression model and Naive Bayes model; Implemented three different kinds of Naive Bayes classifiers: the Gaussian Naive Bayes, the multinomial Naive Bayes, and the Bernoulli Naive Bayes; Compared the performances of the four models on the two distinct datasets: the 20 news group dataset from scikit-learn and the Sentiment140 dataset
Implemented a multilayer perceptron from scratch and analyzed its performance on image classification using the Fashion-MNIST dataset with varying activation function, regularization and normalization methods; built convolutional neural network with PyTorch and tested how different hyperparameter values affect its performance on the Fashion-MNIST dataset
Replicated the work done by An et al. (2020) as a reproducibility challenge; Performed ablation studies
An, Sanghyeon Lee, Minjun Park, Sanglee Yang, Heerin So, Jungmin. (2020). An Ensemble of Simple Convolutional Neural Network Models for MNIST Digit Recognition.