Joshua

This project addresses the challenge of deciphering handwritten mathematical symbols. By leveraging convolutional neural networks (CNNs) and transfer learning, we explored the strengths and weaknesses of different preprocessing steps and neural networks. Our approach employed pre-trained models like Xception and MobileNetV2, which were fine-tuned on a dataset enhanced through normalization and augmentation techniques. These preprocessing steps included standardizing pixel values and using rotations and reflections to improve dataset robustness and reduce overfitting. Our experiments achieved 88% on a small test set, highlighting the potential of CNNs for mathematical notation recognition.See our source code here

BBPI (Bezier and B-Spline Polynomial Interpolation) is a Python library designed for creating, manipulating, and visualizing univariate and bivariate Bezier curves and surfaces. The library provides robust functionality to handle various operations on Bezier curves, such as addition, multiplication, degree elevation, subdivision, and differentiation. Additionally, it supports the construction and evaluation of bivariate Bezier surfaces, making it a versatile tool for applications in computer graphics, CAD, and mathematical modeling. With built-in visualization capabilities using Matplotlib and Plotly, BBPI allows users to easily plot and analyze the shapes and behaviors of these curves and surfaces, making it an invaluable resource for researchers, educators, and developers working with polynomial interpolation and geometric design.

In the FlapAI project, we enhanced the classic Flappy Bird game by introducing new gameplay twists and leveraging artificial intelligence to create an advanced version of the game. Using the Unity game engine and C#, we developed a reimagined version of Flappy Bird that includes a hard mode with new features such as target shooting and obstacle manipulation. Following the game development, we implemented AI algorithms like Proximal Policy Optimization (PPO) and Q-learning using Unity's ML toolkit and machine learning frameworks like PyTorch. These algorithms enabled the AI to autonomously navigate and play the game, adapting to the newly added complexities. Our AI agents, trained with reinforcement learning techniques, demonstrated improved gameplay performance and strategic depth, showcasing the potential of combining game development with machine learning for enhanced user experiences. The project was managed under the advisorship of Dr. Tamer Kahveci.

Tinker is a website that allows users to build their own prompts using variables and different open source models found on HuggingFace. Users can craft their own prompts and store their them for later use or to just use the same prompt with different variables. Whether for text generation or audio to text generation, outputs are stored within each query so they can be easily found later.