News
DeepMind Offers Mathematical Dataset for Training Reasoning in Neural Models
London-based and Google-owned artificial intelligence (AI) research and solution provider DeepMind today released a "large-scale extendable dataset of mathematical questions for training (and evaluating the abilities of) neural models that can reason algebraically," according to the tweet announcing the release.
The dataset can be found on GitHub here, and a related paper on how the dataset can be used to evaluate and train reasoning in neural models can be found here.
According to the dataset's documentation, the code generates "mathematical question and answer pairs from a range of question types...[at] roughly school-level difficulty."
The current version, 1.0, contains 2 million question and answer pairs per model, with questions types labeled as easy, medium or hard so that training can be leveled or mixed.
Some of the many types of mathematical problems included in the dataset are algebra (sequences, linear equations, polynomial roots), calculus (differentiation), measurement (conversion, time-related), and basic arithmetic (mixed expressions, pairwise operations).
Setup instructions for working with the dataset with Python 3 can be found here.
About the Author
Becky Nagel serves as vice president of AI for 1105 Media specializing in developing media, events and training for companies around AI and generative AI technology. She also regularly writes and reports on AI news, and is the founding editor of PureAI.com. She's the author of "ChatGPT Prompt 101 Guide for Business Users" and other popular AI resources with a real-world business perspective. She regularly speaks, writes and develops content around AI, generative AI and other business tech. Find her on X/Twitter @beckynagel.