Annotation
Labeling or structuring raw data, such as tagging objects in images or summarizing text, so machine learning models can learn from it.
As defined in Alexandr Wang - A Primer on AI - [Invest Like the Best, EP. 272]
Where it comes up · 1
Most important claims · 5 of 5Search all claims for “Annotation” →
- Building an AI model involves collecting large amounts of raw data, annotating it into structured form, training the model, and deploying it, and because models are brittle, this must be a continuous loop of recollecting, reannotating, and retraining rather than a one-time build.Alexandr Wang - A Primer on AI - [Invest Like the Best, EP. 272]
- Annotation converts large pools of unstructured data, such as video of pedestrians or long text, into structured labels that algorithms can learn from, and the work is moving from manual labor toward automation that pre-labels data and flags human errors.Alexandr Wang - A Primer on AI - [Invest Like the Best, EP. 272]
- In traditional software the lifeblood is code that dictates system behavior, whereas in AI and machine learning the lifeblood is data, which is the ingredient that determines what the system learns and does.Alexandr Wang - A Primer on AI - [Invest Like the Best, EP. 272]
- Data annotation has historically been done in a low-tech, manually intensive way because most technologists avoid operational messiness, choosing computer science for its more constrained problems.Alexandr Wang - A Primer on AI - [Invest Like the Best, EP. 272]
- The 'dirty jobs' of AI data work involve converting data formats that machines cannot read, such as transaction records and audio intent, into formats that machine systems can use, and these tasks are the hidden foundation of AI.Alexandr Wang - A Primer on AI - [Invest Like the Best, EP. 272]