What is training data?
Quick answer
Training data is the collection of examples a model learns from. Its size, balance and quality set the ceiling on how well the finished model can perform.
Students collect, label and document their own dataset, including a short note on what it fails to represent.
In context
Every property of a model traces back to its data. Gaps in the data become gaps in the predictions, and no amount of clever architecture fixes a dataset that never represented the people it will be used on.
Collecting a dataset is one of the most instructive things a student can do, because the trade-offs become obvious the moment they have to label a thousand items themselves.
In a NASCA classroom
Students collect, label and document their own dataset, including a short note on what it fails to represent.
Related terms
Machine learning
Machine learning is the part of AI where a model improves at a task by being shown labelled examples, rather than by being programmed with explicit rules.
Model bias
Model bias is a systematic error that makes an AI system perform worse for some groups than others, usually because the training data under-represented them.
Student data privacy
Student data privacy covers how a school and its partners collect, store, share and delete information about children, under frameworks such as India's DPDP Act, the GDPR and FERPA.