Researchers At MIT Developed A Machine Learning Model That Can Answer University-Level Mathematics Problems In A Few Seconds At A Human Level

Contrary to humans, machine learning models find it incredibly challenging to handle problems involving differential equations, linear algebra, and multivariable calculus. Even the most advanced models can only answer math problems at the elementary or high school level, and they do not always come up with the correct answers. An MIT multidisciplinary research team has created a neural network model that can quickly and accurately answer college-level arithmetic problems. The model may also automatically explain solutions in university math courses and quickly produce new issues. University students were then given the computer-generated questions to test, and they could not determine whether an algorithm or a human-produced the questions. The study has also been published in the National Academy of Sciences Proceedings. 

Researchers believe their work can be utilized to expedite the creation of course content for extensive residential courses and massive open online courses (MOOCs) with thousands of students. The program could also serve as an automated tutor that demonstrates to pupils how to solve problems in college mathematics. The team believes that by helping teachers to comprehend the connection between courses and their prerequisites, their approach has the potential to enhance higher education. For more than two years, the model has been steadily evolving. In the beginning, the researchers saw that models pretrained using only text could not provide a high accuracy on high school math problems. In contrast, those employing graph neural networks could but would require more extended training periods.

The scientists then experienced a “eureka” moment. They used program synthesis and few-shot learning to convert questions from famous universities’ undergraduate math courses that the model had never encountered before into programming tasks. The researchers added a further stage of “fine-tuning” before feeding these programming tasks to a neural network. The employed pre-trained neural network, Codex, was “fine-tuned” on both text and code. The pretrained model was trained on data containing millions of lines of code and natural language words, allowing it to understand the link between text and code. With just a few question-code examples, the model can now convert a text question into code and then run the code to provide an answer because it can recognize different relationships between text and code. This method showed an enormous improvement in accuracy—from 8 to 80 percent—. By giving the neural network a set of arithmetic problems on a subject and then asking it to come up with a new challenge, the researchers also utilized their model to generate queries. We also examined these computer-generated questions by displaying them to college students. Students gave both human- and machine-generated questions comparable marks for the level of difficulty and suitability for the course because they could not distinguish between the questions produced by a human or an algorithm.

The team makes it clear that their effort aims to pave the way for people to begin using machine learning to solve more challenging problems rather than to replace human professors. Although the team is delighted with the results of their strategy, there are several drawbacks that they must overcome. Due to computational complexity, the model cannot answer questions with a visual component and cannot resolve computationally intractable issues. Along with getting past these obstacles, they want to build the model up to hundreds of courses so that it can improve automation and offer insights into course design and curriculum.

This Article is written as a research summary article by Marktechpost Staff based on the research paper 'A neural network solves, explains, and generates university math problems by program synthesis and few-shot learning at human level'. All Credit For This Research Goes To Researchers on This Project. Checkout the paper and reference article.

Please Don't Forget To Join Our ML Subreddit

Khushboo Gupta is a consulting intern at MarktechPost. She is currently pursuing her B.Tech from the Indian Institute of Technology(IIT), Goa. She is passionate about the fields of Machine Learning, Natural Language Processing and Web Development. She enjoys learning more about the technical field by participating in several challenges.

🚀 LLMWare Launches SLIMs: Small Specialized Function-Calling Models for Multi-Step Automation [Check out all the models]