I deployed a reinforcement learning agent to complete multiple levels of a platformer game. By implementing a genetic algorithm, I enhanced the neural network’s performance, while utilizing sightlines to enable precise obstacle reactions, significantly speeding up level completion.

The project introduces an advanced supervised learning algorithm that skillfully employs clustering techniques, specifically Gaussian Mixture Models (GMM), to pinpoint hotspots within each class, significantly enhancing model interpretability. By integrating Expectation-Maximization (EM) for optimal parameter estimation and employing a robust bootstrapping algorithm for model selection, this method not only achieves a notable reduction in prediction costs but also provides deep probabilistic insights. These insights give a clearer understanding of the data's inherent distributions, setting a new benchmark for efficiency and analytical depth in predictive modeling.

This project analyzes IMDB reviews to classify sentiments as positive or negative. Through preprocessing steps like data cleaning and TF-IDF, followed by dimensionality reduction with PCA, various machine learning models were evaluated, including Logistic Regression, KNN, LDA, QDA, and Random Forests. The study meticulously assessed model performance, revealing key insights into predictive accuracy and model selection for sentiment analysis, thereby providing a comprehensive approach to understanding and improving sentiment classification techniques.

In this project, the players find themselves in a 2D arena game where they have to survive against an array of killer cyborgs controlled via a partially damaged transmitter. The game intricately blends action and strategy, challenging players to use walls to their advantage, and smartly command cyborgs to outmaneuver threats. With its unique blend of character-based graphics and interactive gameplay, the project offers an immersive experience.