cnn-early-layers-encode-lowlevel-features-typically-frozen

IN premise — summaries/2026/08/24/wiki-Fine-tuning_deep_learning.md

Created 2026-08-24T17:11:09+00:00

In CNN fine-tuning, early (input-adjacent) layers encode low-level features such as edges and textures and are typically frozen, while later layers encode task-specific high-level features and are fine-tuned.

Summary

When adapting a pre-trained convolutional network to a new task, the standard practice is to leave the layers that detect basic visual patterns like edges and textures untouched, and only update the deeper layers that capture more abstract, task-relevant structure. This matters because it keeps the low-level visual vocabulary stable while letting the model reshape its higher-level interpretation, which reduces training cost and avoids destroying features that are broadly useful across tasks.