Idea Ripper // ArchiveCards: 1555 [Live] | Updated: 2026-10-14
Card #c84IR // #c84Synthesis

Historically, the power of machine learning algorithms has increased over time, largely due to the growth in the size and complexity of the models.

— In our words · Source: Tiny Machine Learning QuickStart (Simone Salerno)

[Use when]

Use when a roadmap assumes bigger models are always better.

[Why it matters]

Progress-by-scale collides with devices that cannot scale.

More from Tiny Machine Learning QuickStart

#c80This convergence has given birth to Tiny Machine Learning (often shortened as TinyML®, trademark of the EdgeAI foundation)—a field that…Use when "edge AI" is used as a buzzword without naming the constraint. #c81Unlike cloud-based machine learning systems with virtually unlimited resources, TinyML development requires careful optimization to fit…Use when a team ports a cloud model and is shocked it will not fit. #c82Due to size, efficiency, power, and cost constraints, embedded hardware is characterized by limited resources (CPU, program space, and RAM).Use when someone argues "just use a bigger chip" as if cost and power were free.