What would happen if K in KNN is dynamically learned instead of fixed?

It becomes an adaptive model that adjusts locality based on density.

Instead of fixed K, the model can choose K based on local density—smaller K in dense regions and larger K in sparse regions. This improves adaptability but increases complexity and instability if not regularized.