Overfitting

noun

Overfitting occurs when a machine learning model becomes overly complex, capturing noise and outliers in its training data rather than the underlying patterns. This happens because the model tries to minimize training errors at the expense of performance on unseen data.

For example, a language model trained on a specific dataset might memorize quirks or errors in that data, leading to incorrect predictions when faced with new inputs. Overfitting is common in models with too many parameters relative to the amount of training data.

For developers, overfitting means wasted computational resources and unreliable outputs. It forces them to use techniques like regularization, cross-validation, or simpler models to ensure the AI works well in real-world scenarios. A poorly trained image generator might produce distorted outputs if it overfits to training images with artifacts.

A chatbot trained on a small dataset of customer service scripts may overfit and fail to answer questions outside those scripts.

Découvrez nos autres services

ElpisIA, c’est tout un univers — explorez nos autres services.