Marc-Antoine Nadeau
Home CV Projects Interests Contact
Artificial Intelligence Projects
Trading Deep Q-Networks (TDQN) - A Reinforcement Learning Approach to Trading

Trading Deep Q-Networks (TDQN) - A Reinforcement Learning Approach to Trading
Reinforcement Learning

Enhanced a Trading Deep Q-Network (TDQN) based on academic research by integrating Rainbow DQN components (Dueling Architecture, Noisy Nets, PER) and implementing an Actor-Critic algorithm. Achieved improved performance on Tesla and Apple stocks, with consistent profitability even during market downturns.

reinforcement-learning deep-q-network trading actor-critic
Deep Q-Network and Expected SARSA for Atari Games and Robotic Control (Acrobot)

Deep Q-Network and Expected SARSA for Atari Games and Robotic Control (Acrobot)
Reinforcement Learning

Applied Deep Q-Network (DQN) and Expected SARSA algorithms to classic reinforcement learning benchmarks, including an Atari game and the Acrobot robotic arm system, demonstrating effective learning in both visual and control-based environments.

reinforcement-learning deep-q-network expected-sarsa atari acrobot
Performance Evaluation of Bandit Algorithms on a Bernoulli Bandit Problem

Performance Evaluation of Bandit Algorithms on a Bernoulli Bandit Problem
Reinforcement Learning

Investigated the performance of the Q-Network algorithm on a Bernoulli Bandit problem, focusing on its ability to balance exploration and exploitation.

reinforcement-learning bandit-algorithms q-network
Comparative Analysis of LLMs and Traditional Models for Emotion Classification on the GoEmotions Dataset

Comparative Analysis of LLMs and Traditional Models for Emotion Classification on the GoEmotions Dataset
Large Language Models

This study evaluated DistilGPT-2, GPT-2, BERT, Multinomial Naive Bayes, and Random Forest models on the GoEmotions dataset, focusing on accuracy, recall, and efficiency. Multinomial Naive Bayes showed moderate performance, leveraging Count Vectorization and Laplace smoothing effectively for text processing, but class imbalance led to lower recall for less frequent emotion labels.

large-language-models emotion-classification goemotions
Comparing MLPs, CNNs, and Pre-trained MobileNetV2 for Organ Classification on the OrganAMNIST Dataset

Comparing MLPs, CNNs, and Pre-trained MobileNetV2 for Organ Classification on the OrganAMNIST Dataset
Computer Vision

The complexity of modern medical imaging datasets has driven the need for advanced machine learning models capable of achieving accurate and efficient classification. This repository explores the OrganAMNIST dataset, a challenging benchmark derived from medical imagery, to evaluate the effectiveness of different neural network architectures for organ classification.

computer-vision image-classification organ-amnist
Diabetes Diagnosis from Health Indicator Data Using Logistic Regression

Diabetes Diagnosis from Health Indicator Data Using Logistic Regression
Logistic Regression

We diagnosed diabetes using the Diabetes Health Indicators dataset. Our preprocessing involved handling class imbalance, cleaning and scaling data, and removing redundant features to enhance model performance and ensure robust predictions.

logistic-regression diabetes-prediction health-indicators
Predicting Oral Temperature Using Linear Regression on Infrared Thermography Data

Predicting Oral Temperature Using Linear Regression on Infrared Thermography Data
Linear Regression

We predicted average oral temperature using the Infrared Thermography Temperature dataset. Key preprocessing steps included cleaning missing and duplicate data, scaling features, and removing highly correlated variables to improve model reliability and accuracy.

linear-regression temperature-prediction infrared-thermography
Numerical Methods Projects
Fourier Transform Toolkit for Image Processing

Fourier Transform Toolkit for Image Processing
Signal Processing

A comprehensive toolkit for performing 1D and 2D Discrete Fourier Transforms (DFTs) and Fast Fourier Transforms (FFTs) on images. Features include intuitive visualizations, image denoising, and compression through frequency domain filtering. Additionally, it offers runtime benchmarking to compare naive and FFT algorithms, along with validation of inverse FFT (IFFT) results.

signal-processing fourier-transform fft image-processing
Monte Carlo and Gauss-Legendre Numerical Integration Methods Using MATLAB

Monte Carlo and Gauss-Legendre Numerical Integration Methods Using MATLAB
Numerical Integration

This repository contains implementations of two numerical integration techniques using MATLAB: Monte Carlo Integration and Gauss-Legendre Quadrature.

gauss-legendre monte-carlo numerical integration matlab
Networking Projects
Command-Line DNS Client for A, NS, and MX Record Queries

Command-Line DNS Client for A, NS, and MX Record Queries
Networking

Implements a DNS client that performs A, NS, and MX record queries using UDP sockets. Includes retransmission logic, detailed response parsing, and command-line interface with support for flags like timeout, retries, and port override.

dns networking sockets udp
Systems Programming Projects
OS Shell Simulation: Command Parsing, Scheduling, and Demand Paging

OS Shell Simulation: Command Parsing, Scheduling, and Demand Paging
Operating Systems

A fully simulated OS shell developed from the ground up. Features include command parsing, script execution, multiple process scheduling algorithms (FCFS, SJF, Round Robin, AGING), and demand paging using LRU replacement. Supports multi-process management with robust memory isolation.

operating-systems shell scheduling demand-paging
Conway's Game of Life in ARM Assembly

Conway's Game of Life in ARM Assembly
ARM Assembly

Implemented Conway's Game of Life on an ARM DE1-SoC machine in Assembly.

assembly game-of-life conway
Full-Stack Projects
FRED - Full-Stack Web Application for Managing Personal Finances

FRED - Full-Stack Web Application for Managing Personal Finances
Full-Stack Application

FRED is a full-stack application designed to help users manage their personal finances. It features a user-friendly interface for tracking expenses, incomes, and budgets, with advanced functionalities powered by ChatGPT for personalized financial advice.

full-stack typescript personal-finance chatgpt-integration
SportCenter - Full-Stack Sports Management Application

SportCenter - Full-Stack Sports Management Application
Full-Stack Application

SportCenter is an application that enables owners and instructors to manage fitness class schedules, including creating, editing, and canceling classes. Customers can view, register for, and cancel classes, as well as track past and upcoming sessions. An email system confirms reservations.

full-stack sports-management spring boot java
AssetPlus Application - A Full-Stack Hotel Management System

AssetPlus Application - A Full-Stack Hotel Management System
Full-Stack Application

AssetPlus is a desktop application designed to assist hotels in managing their assets efficiently. This application streamlines the tracking, maintenance, and reporting of hotel assets, ensuring optimal utilization and management.

full-stack hotel-management java