Glossary term
Glossary term
Training and Fine-Tuning
Any mathematical transform or technique that shifts the range of a label, a feature value, or both. Some forms of scaling are very useful for transformations like normalization.
Common forms of scaling useful in Machine Learning include:
linear scaling, which typically uses a combination of subtraction and division to replace the original value with a number between -1 and +1 or between 0 and 1.
logarithmic scaling, which replaces the original value with its logarithm.
Z-score normalization, which replaces the original value with a floating-point value representing the number of standard deviations from that feature's mean.
Created for this library
A research team studies scaling laws between compute, data, and parameter count to plan its next-generation model training budget.
An ML platform team plans capacity using scaling-law intuition rather than just trial and error.
A startup picks model size and dataset size based on scaling-law guidance to balance cost and quality.
Definition source: Google for Developers Machine Learning Glossary | Creative Commons Attribution 4.0 License