Researchers show supervised learning may not be fundamentally different from unsupervised

A new method allows machine learning models to select their internal parameters without ever seeing the target labels (y values), then adds those labels afterward without changing anything—challenging the idea that supervised and unsupervised learning are categorically different. The finding suggests that what makes learning "supervised" might be less about the learning process itself and more about what you do with the results at the end.

If supervised and unsupervised learning are less fundamentally different than decades of machine learning theory has assumed, it could reshape how researchers think about model design, potentially opening new ways to use unlabeled data or reduce dependence on labeled training sets.