M31.3 CONNECT THE MECHANISM
Reuse learning when the next task or domain changes
A solar farm has 600 photos of cracked panels, too few to train a vision model. Borrow one trained on a million images, then keep it working when the weather changes.
LESSON OVERVIEW14 min lesson
Lesson overview
A solar farm has 600 photos of cracked panels, too few to train a vision model. Borrow one trained on a million images, then keep it working when the weather changes.
What you’ll explore
- Choose between feature extraction and fine-tuning, estimate target performance under covariate shift with importance weights, explain domain-adversarial training, and check for negative transfer on labeled target data.
GO TO THE SOURCE
Original explanations, connected to the research.
How transferable are features in deep neural networks? (Yosinski et al., 2014)Domain-Adversarial Training of Neural Networks (Ganin et al., 2016)Dive into Deep Learning — Fine-TuningDive into Deep Learning — Environment and Distribution Shift (covariate shift correction)Suggest a correction
A precise note can make an explanation better.
Choose the scene and describe what needs attention. Download a feedback file to share through a channel you already use. This page does not send feedback or connect you with a reviewer.