Chapter 09 · Sustainability & AISustainability & AI

Transfer learning

Meaning statusEstablishedSource recordDirect document linkedWhy these are different

Definition

Reusing a model trained on one task or domain as the starting point for another, so that less data and computation are needed for the new task.

References

IEEE TKDE 2010Pan & Yang — A Survey on Transfer Learning

This source provides part of the technical or institutional basis for the definition.

Stanford CRFM / arXivBommasani et al. — On the Opportunities and Risks of Foundation Models (2021)

This source supports the explanation of how the term is applied, measured or governed in practice.

Overview

What it means

A model pre-trained on broad data — images, text, spectra — already encodes general patterns; fine-tuning adapts them to a specific problem. Pan and Yang's 2010 survey systematised the field. Transfer learning is why small teams can build strong tools on top of large pre-trained models.

How it is used

In environmental work it is ubiquitous: adapting general vision models to a particular ecosystem's camera-trap images, or general language models to sustainability-report analysis, without training from scratch.

Why it matters

Transfer learning is the main mechanism by which expensive, centralised model training propagates capability — and its errors — into thousands of downstream uses. Efficiency gains and concentration of influence travel together.