Transfer Learning Explainer
Transfer learning starts with a model trained on one problem and reuses part of what it learned for another. Instead of training every parameter from scratch, developers use pretrained representations as features or adapt them with a smaller task-specific dataset.
A common workflow freezes the base model, adds a new output layer, and trains that layer on the target data. Developers may then unfreeze some pretrained layers and fine-tune them with a low learning rate so useful general features adjust without being overwritten abruptly.
Transfer learning can reduce compute and labeled-data requirements, especially when the source and target tasks share structure. Image, language, audio, and scientific models often serve as reusable starting points for specialized applications.
Transferred knowledge can be mismatched or harmful. A base model may carry bias, security weaknesses, licensing constraints, or features irrelevant to the new domain. Evaluation must use target-domain examples and compare against simpler baselines rather than assuming pretraining always helps.
Transfer learning reuses representations learned for one task or dataset as a starting point for a related new task.
Key terms include pretrained model, source task, target task, feature extraction, frozen layer, fine-tuning, learning rate and domain shift.
Feature extraction keeps pretrained layers fixed and trains a new head. This is efficient and protects the existing representation during early adaptation.
Fine-tuning updates some or all pretrained weights after the new head has stabilized, typically using smaller updates to avoid destroying useful features.
Bias, memorized information, brittle shortcuts, and vulnerabilities can move with the pretrained model as easily as useful representations.
Teams should document the source model, data assumptions and license, test the actual target population, and monitor whether adaptation degrades earlier safeguards.
Explore more "Explainers"
Discover additional explainers across politics, science, business, technology, and other fields. Each explainer breaks down a complex idea into clear, everyday language—helping you better understand how major concepts, systems, and debates shape the world around us.