noun
Model fine-tuning is the process of taking a pre-trained artificial intelligence model—like a language or image generator—and tweaking it to perform better on a specific task. Instead of training a model from scratch, which is expensive and time-consuming, fine-tuning leverages existing knowledge and refines it with new, task-specific data.
For example, a general-purpose AI image generator might be fine-tuned on medical scans to improve its accuracy in detecting tumors. The process typically involves feeding the model a smaller, specialized dataset and adjusting its parameters to reduce errors. This can cost significantly less than full training but still requires computational resources and expertise.
For users, fine-tuning changes what’s possible without needing to build a model from scratch. It allows customization for niche applications, like a business fine-tuning a chatbot to handle customer service for a specific product. However, it also introduces costs—both in money and time—and may require technical skills to implement effectively.
A startup fine-tunes a language model on customer support transcripts to improve its responses to common complaints.
In other languages
See also: Transfer Learning · Hyperparameter tuning · Overfitting