Research Internship at CKI - Abstract Reasoning, Math problem solving

Position: Research Internship at CKI

Project: Abstract Reasoning, Math problem solving

Scientific supervisor: Prof. Dr. Andrey Ustyuzhanin

Introduction to CKI:

The Constructor Knowledge Institute (CKI) intends to set the worldwide standard for research into Computer Science, AI and Machine Learning, Robotics, and Neuroscience, operating in strong contact with industry, and:

• Leveraging CS technologies to address challenges in various fields, delivering innovative solutions tailored to industry needs.
• Providing research opportunities, mentorship, and involvement in collaborative projects to young researchers and PhDs.
• Encouraging interdisciplinary research by fostering collaboration between diverse fields, emphasizing the integration of theoretical research with practical application

Project: Abstract Reasoning, Math problem solving

The goal of this project is to design and train an architecture to tackle reasoning problems akin to the Abstract Reasoning Challenge (ARC) – a set of tasks designed to evaluate machine reasoning abilities in abstract problem-solving. The AI model will be trained to understand, analyze, and solve reasoning tasks (from the ARC challenge and alike), aiming to achieve state-of-the-art performance in abstract reasoning. The project will explore the AI’s potential in handling complex, logic-based reasoning problems that require pattern recognition, inference, and generalization beyond simple memorization.

Challenges / key research questions:

• Handle the complexity and diversity of abstract reasoning tasks.
• Ensure the AI model can generalize beyond memorization to solve new, unseen problems.
• Develop robust inference/sample test-time techniques that mimic human-like reasoning.
• Balance computational efficiency with model performance on large-scale tasks.

Requirements:

• Python skills
• Experience with AI libraries
• Prompt engineering
• Math and data processing skills are plus