NH-LoRA: Future-Aware Structural Expansion of Low-Rank Adapters for Rehearsal-Free Class-Incremental Learning
- A future-aware continual learning method that treats LoRA capacity as a structural decision, not a fixed adapter budget.
PyTorchNeuromorphicCILViTArchitecture Experiment
About
NH-LoRA is a rehearsal-free class-incremental learning method built on a frozen Vision Transformer and parameter-efficient low-rank adaptation. It separates reusable cross-task knowledge from task-specific expandable memory, then uses a task-aware planner to control structural decisions such as reuse, rank expansion, or new-slot creation. During learning, sparse routing and post-task consolidation help balance plasticity and stability while keeping parameter growth under control.