prelearned

Prelearned is an adjective derived from the prefix pre- meaning before, combined with learned, and is used to describe knowledge, representations, or models that have already been learned prior to a current task. In general usage, prelearned knowledge exists before new learning takes place and can influence how new information is acquired or how problems are approached.

In modern computing, the term is closely related to pretrained or prelearned models and features. A prelearned model is one that has been trained on a large dataset and is then adapted for a specific downstream task, typically through fine-tuning. This approach contrasts with training a model from scratch and often reduces training time, data requirements, and computational cost. Prelearned representations, such as word embeddings or deep neural network features, are used to provide a ready-made basis for new tasks.

Common examples include word embeddings like Word2Vec or GloVe, image classifiers pretrained on ImageNet used as feature extractors, and large language models that are pretrained on broad corpora before task-specific fine-tuning. Limitations of prelearned approaches include domain mismatch, where the prelearned knowledge does not align with the target domain, and the risk of inheriting biases or errors from the pretraining data.

In cognitive science, the term can describe knowledge that individuals acquire before engaging in a task, which then informs subsequent learning or decision-making.