20-40 samples can find significant improvements in 10,000+ examples. Wanna know how?

Many organizations lack the resources to build massive models with trillions of parameters trained on vast datasets. Hence, for trustworthy insights, we must build small if we can’t build big.

Micro models, learned from very few data points, speeds up the process of modeling. It also simplifies AI tools, enabling rapid iteration and easier interpretation, unlocking significant value from minimal data.

Micro models, offers several benefits:

• They allow quick interaction with subject matter experts to build a test suite of highly informative examples.
• Good micro model results avoid big model issues like validation, reproducibility, understandability, high CPU costs, privacy concerns, and poor performance on specific problems.
• Poor micro model can be used to justify the use of larger models.
• Good micro models can replace failing general models.
• Together, micro models can enhance the output of larger models.

1 Contents

• Motivation
• Intro to the code
• [Open issues[(what2do.html)

Work in progress:

• Install instructions
• How Much Data Do we Need?