Projects
Coreset Algorithms for Clustering and Streaming
Python implementation of Coreset algorithms for clustering and streaming. The code in this repository can be used to generate Coresets for use-cases such as median estimation, minumum-enclosing ball (MEB), k-center clustering and Gaussian mixture models (GMMs).
Settlers of Catan AI Framework
Settlers of Catan boardgame built in Python. The goal of this project is to implement full multiplayer game functionality and use machine learning to build an AI player that can effectively explore-exploit heuristic strategies.
References:
Xenou, Konstantia, Georgios Chalkiadakis, and Stergos Afantenos. "Deep Reinforcement Learning in Strategic Board Game Environments." European Conference on Multi-Agent Systems. Springer, Cham, 2018.
Gendre, Quentin, and Tomoyuki Kaneko. "Playing Catan with Cross-Dimensional Neural Network." International Conference on Neural Information Processing. Springer, Cham, 2020.
Tennis Player Performance Prediction
Under the guidance of Prof. Jiebo Luo, I used efficient feature extraction on historical tennis match data, combined with a machine learning implementation in Python to predict the likelihood of professional tennis player success with 80% accuracy. I implemented code to create player-specific statistical feature sets aggregated from individual match data, and used neural network and logistic regression classification models to categorize player success.
Adversarial Search: Ultimate Tic Tac Toe
An Ultimate Tic Tac Toe framework in Java, with an implementation of adversarial search using MiniMax with Alpha-Beta pruning. Ultimate Tic Tac Toe comprises nine 3x3 Tic Tac Toe boards, and the goal is to win 3 boards. I also developed a heuristic AI player, which was tested to beat a control player in 99 out of 100 games.
German WWII Enigma Machine
A complete implementation of the Enigma Machine in Python. I use an object-oriented framework to design the plugboard, reflector and rotor set, allowing for full encryption and decryption functionality. I also implement code to crack the Enigma cipher, using a known-plaintext attack methodology.
Non-Linear Dynamics of the Damped and Driven Pendulum
I investigated the non-linear dynamics of the damped and driven pendulum and developed a theoretical framework for the system. I then computationally solved the classical mechanics problem using Mathematica to discover regions of chaotic and non-chaotic motion.
Augmented Audio Reality: Binaural Headphones
Designed, built and tested binaural headphones with real-time recording and filtering capability, with a 12ms latency. Modeled the head-related tranfer function (HRTF) using a Neumann head microphone, and implemented the FFT algorithm in C++ to enable real-time audio filtering.
Brownian Motion Stock Price Evolution
A statistical framework in Python to predict stock price evolution using geometric Brownian motion. The model was tested to have under 5% error using Monte Carlo simulations on 2 years of historical Nike stock prices.
Reference: Dmouj, Abdelmoula. "Stock price modelling: Theory and Practice." Masters Degree Thesis, Vrije Universiteit (2006).