Curriculum (data)
A training strategy that artificially orders the presentation of data, starting with the simplest examples and gradually increasing the difficulty over time.
Think of It Like This
Like teaching a child basic addition and subtraction before suddenly dropping a calculus textbook on their desk.
Inspired by how humans learn, curriculum learning helps a model find a good initial set of weights before exposing it to complex, noisy outliers that might derail optimization. While it makes intuitive sense, it requires significant manual effort to score and sort the training data. In practice, modern deep learning often favors massive, randomized data ingestion instead of carefully curated curricula.