Our Projects

Neuro-Gradient is a newly emerging initiative that bridges computational neuroscience, cognitive science, and deep learning. The section will focus on modeling processes such as reasoning, memory, and forgetting using mathematics and code. Planned activities include hackathon participation, workshops (e.g., EEG signal analysis), and collaborations with other student research groups. While the section is still in its early stage, it aims to grow into a hub for neuro-AI research within the club.

Team:

The aim of this project is to develop an intelligent Discord chatbot for the Gradient Science Club server that streamlines communication and club management. The chatbot will support both members and the board by providing: - Information about upcoming and past events. - A database of news, projects, and club structure. - Ready-to-use administrative and management commands. - Integration with external tools through API support. The project also includes a custom framework for evaluating the chatbot, focusing on: - Manual quality metrics for responses. - Retrieval-Augmented Generation (RAG) tests, specifically assessing the correctness of context-based answers. Why You Should Join: By participating, you will gain hands-on experience with modern chatbot technologies, work with APIs and RAG systems, collaborate in a multidisciplinary team, and contribute to a tool that directly supports the life of the club.

Team:

SD

This section focuses on projects that use reinforcement learning. The current project aims to develop an agent that learns to play a simple computer game and then learns to play it “in its imagination.”

Team:

The project focuses on classifying the features and species of fruit flies from images. It is carried out in collaboration with the Genetics Club at the University of Gdańsk.

Team:

A project focused on extracting text data from images of identity documents using Optical Character Recognition (OCR). The work included image preprocessing (enhancement, denoising) and deep learning-based OCR models to improve recognition accuracy. The project has now been completed and provided valuable experience in building document analysis pipelines.

This project demonstrates the use of AI to transform images into different artistic styles using Stable Diffusion and custom-trained style models. Leveraging techniques from Dream Styler, the project allows users to apply a wide variety of visual styles to their images while preserving content features, effectively blending creativity and AI. Key Highlights: - Custom style models trained on multiple images for diverse artistic effects. - Transfers style onto new images while maintaining the original content. - Supports generating style grids for previewing multiple transformations at once. - Explores advanced AI techniques in text-to-image generation and style transfer. - Demonstrates practical applications of neural networks in digital art and creative AI.

RL Karting is a completed project where autonomous AI agents were trained to drive karts in a Unity-based racing environment. Built on the Unity Karting Microgame, the project demonstrates how reinforcement learning can be applied to teach agents to navigate tracks, avoid obstacles, and compete in races without human input. The project showcases fully trained neural network models that exhibit intelligent driving behaviors, adapting to different tracks and scenarios. It serves as an engaging example of combining game development, machine learning, and AI research in a single interactive environment.