Dataset
Level 1
A collection of examples gathered for a model to learn from or be tested on.
Term 1 of 7 in Data and databases
In plain language
A dataset is the raw material of machine learning: the emails, photos, transactions or documents that teach a model what normal looks like. Its quality caps everything downstream.
Think of it like this
The ingredients. No technique rescues a dish built on spoiled ingredients.
Why it matters
If a dataset under-represents a group or a situation, the model will quietly be worse at it. Most AI bias starts here.
Next terms in Data and databases
- Structured and unstructured dataStructured data lives in neat rows and columns; unstructured data does not.
- EmbeddingA list of numbers that captures the meaning of a piece of text or an image.
- Vector databaseA database built to store embeddings and find the closest matches fast.
- Retrieval augmented generationFetching relevant documents first, then asking the model to answer using them.