Glossary term
Glossary term
Training and Fine-Tuning
A graph of weight(s) versus loss. Gradient descent aims to find the weight(s) for which the loss surface is at a local minimum.
Created for this library
An ML researcher visualizes the loss surface around the final checkpoint to study how flat or sharp the converged solution is.
A research team studies the loss surface of large models to understand why some optimizers reach better solutions than others.
An ML platform team uses loss-surface visualizations during release reviews to communicate training stability to non-technical reviewers.
Definition source: Google for Developers Machine Learning Glossary | Creative Commons Attribution 4.0 License