AI RESEARCH

Vision Transformers that Never Stop Learning

arXiv CS.LG

ArXi:2603.07787v1 Announce Type: new Loss of plasticity refers to the progressive inability of a model to adapt to new tasks and poses a fundamental challenge for continual learning. While this phenomenon has been extensively studied in homogeneous neural architectures, such as multilayer perceptrons, its mechanisms in structurally heterogeneous, attention-based models such as Vision Transformers (ViTs) remain underexplored.