LoRARecycleAlgorithm<T, TInput, TOutput>
Models & Types · AiDotNet.MetaLearning.Algorithms
Implementation of LoRA-Recycle (Hu et al., CVPR 2025).
Inheritance: ModelBase<T, TInput, TOutput> → MetaLearnerBase<T, TInput, TOutput> → LoRARecycleAlgorithm<T, TInput, TOutput>
Implements: IGeneratedParameterRegistrar<T>
How It Works
LoRA-Recycle enables tuning-free few-shot adaptation by recycling pre-tuned LoRA adapters. It maintains a bank of task-specific LoRA adapters and a prototype-based selection mechanism that computes task embeddings from the support set to weight and combine stored adapters without any gradient-based inner-loop optimization.
Algorithm:
Key difference from MAML: No gradient-based inner-loop optimization at inference time. Adaptation is a single forward pass through the prototype encoder followed by adapter selection and fusion.
Constructors
| Constructor | Summary |
|---|---|
LoRARecycleAlgorithm(LoRARecycleOptions<T, TInput, TOutput> options) |
Properties
| Property | Summary |
|---|---|
MetaLearningAlgorithmType AlgorithmType { get; } |
Methods
| Method | Summary |
|---|---|
IModel<TInput, TOutput, ModelMetadata<T>> Adapt(IMetaLearningTask<T, TInput, TOutput> task) |
|
T MetaTrain(TaskBatch<T, TInput, TOutput> taskBatch) |