2021 NLP challenge of a multi-class classification task with 28 classes, where the task is to assign the correct job category to a job description. In this competition, our team secured 8th position in the private leaderboard.
Integrated MLFlow project on twitter social spam detection problem in conda and docker env to perform experiments and access metadata using MLFlow UI
Detailed analysis of 9 different neural network algorithms including Transformer models on sentiment analysis and tweet extraction to understand and report performance of each method on classification task
Generated artificial images using Deep Convolution GANS and denoised images with deep autoencoder for MNIST and CIFAR10 images
Built an ML model for spam detection by applying different case studies for feature selection (sklearn, PCA) and performed analysis to identify outliers and clusters in the twitter spam dataset
Analyzed Waveform dataset by using K-NN algorithm and experimented by generating artificial imbalancy in the labels to test robustness of KNN
Graph Convolutional Networks (GCN) and Graph Attention Networks (GAT) implementation using pytorch geometric on the cora dataset
Graph Neural Network implementation using DGL and benchmark datasets like Cora, Citeseer and Pubmed.
This project involves data cleaning, feature extraction, model building, hyperparameter tuning with cross validation and PCA in R.
Application of Hidden markov models to calculate the probability of future observations using Forward, Backward and Viterbi algorithms.
Generative Adversarial Neural Networks used to generate new images for MNIST and CIFAR10 datasets.
Reinforcement techniques like policy iteration, value iteration, Q-learning are studied and experimented on different OpenAI environments (using the gym python library).
Machine Learning algorithms were built from scratch with and without using Kernel trick. Algorithms like PCA, KMeans, LASVM, OCSVM, Passive aggressive online algorithm were covered.
Deployed ML model using FastAPI to predict optimal keyword bid for digital ad marketing data and performed analysis on how to improve KPIs, identified best performing markets, campaigns and keywords