A city harpsichord workshop has only a few dozen labeled jack stills for a vision model. Training from random weights is too slow for the schedule. What should the candidate do?
Select an answer to reveal the explanation.
Short Explanation
Think of a pretrained vision stack as a seasoned apprentice who already knows edges and textures. You only teach the shop-specific jack parts—far less data and compute than starting from a blank slate.
Full Explanation
Transfer learning starts from a pretrained vision model and adapts it to the small labeled set. That path is the official way to reduce both data and computation versus training from random weights. Kernel writing and cluster scale do not replace that model-side efficiency.