MetaLoRABankAlgorithm<T, TInput, TOutput>
Models & Types · AiDotNet.MetaLearning.Algorithms
Implementation of Meta-LoRA Bank (2024).
Inheritance: ModelBase<T, TInput, TOutput> → MetaLearnerBase<T, TInput, TOutput> → MetaLoRABankAlgorithm<T, TInput, TOutput>
Implements: IGeneratedParameterRegistrar<T>
How It Works
Meta-LoRA Bank maintains a bank of K diverse LoRA modules, each representing a learned adaptation pattern. For a new task, a gating network computes task-conditioned scores and selects the top-K modules via sparse gating. The selected modules are combined with learned gating weights to produce the task-specific adaptation.
Algorithm:
Advantages: Each module specializes in a different type of task adaptation. The gating mechanism enables compositional generalization — novel tasks can be handled by new combinations of existing modules.
Constructors
| Constructor | Summary |
|---|---|
MetaLoRABankAlgorithm(MetaLoRABankOptions<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) |