Transfer learning and fine-tuning

Choose your learning level

BeginnerIntermediateAdvanced

Start with plain-language explanations. Switch levels at any time.

Beginner explanation

Transfer learning starts with a model that has already learned useful patterns. Fine-tuning adjusts some or all of its parameters for a new task. This can reduce the amount of task-specific training needed, but does not remove the need for evaluation.

Intermediate explanation

A frozen feature extractor leaves pretrained weights unchanged while a new prediction head learns. Fine-tuning updates selected weights. Keep validation and test data separate and check whether the original training domain resembles the new task.

Advanced explanation

Parameter-efficient methods update a smaller parameter set, such as low-rank adapters, while retaining most pretrained weights. Compare generalization, forgetting and data leakage, not only training loss. Prompting and retrieval change inputs; fine-tuning changes learned parameters.

PUT IT IN CONTEXT

A worked example

A fictional plant classifier can reuse a pretrained image backbone and learn a head for five plant types. It must still be tested on different specimens and lighting, not just more photos of the same plant.

TRY IT YOURSELF

Practice the idea

Classify these operations: add a manual to the prompt; train an output head; update all network weights.

Your working notes (not saved or sent)

Reveal the answer guide

Adding the manual changes context, not weights. Training only the output head is a frozen-backbone transfer approach. Updating all weights is full fine-tuning. All three require task-specific evaluation.

Go further: advanced challenge

Design a split that prevents photos of the same physical object from appearing in both training and final test sets.

An open-ended discussion task. Explain your assumptions and evidence; there may be more than one defensible answer.

Continue the conversation

What part of this idea would you like to test or question? Compare your reasoning with the community.

Find related forum discussions ↗

Opens a topic search in the existing AstraCruz forum. Results depend on available posts; signing in may be required. No discussion or activity is fabricated.

Browse all forums