DPGNAlgorithm<T, TInput, TOutput>

Models & Types · AiDotNet.MetaLearning.Algorithms

Implementation of DPGN (Distribution Propagation Graph Network) (Yang et al., CVPR 2020).

Inheritance: ModelBase<T, TInput, TOutput> → MetaLearnerBase<T, TInput, TOutput> → DPGNAlgorithm<T, TInput, TOutput>

Implements: IGeneratedParameterRegistrar<T>

For Beginners

DPGN uses TWO graphs working together:

Graph 1 - Point Graph:

  • Nodes = examples (support + query)
  • Edges = feature similarity
  • Message passing: Share feature information between similar examples
  • Result: Refined, context-aware features

Graph 2 - Distribution Graph:

  • Same nodes but edges = distribution similarity
  • Passes uncertainty/confidence information
  • Result: Each node knows how certain it is

Why two graphs? Knowing features isn’t enough. You also need to know CONFIDENCE. Two examples might have similar features but very different confidences. The distribution graph captures this distinction.

How they interact: After each propagation layer:

  1. Point graph updates features (making them more discriminative)
  2. Distribution graph updates confidences (making uncertainty estimates better)
  3. Both use each other’s output as input for the next layer

How It Works

DPGN constructs dual graphs over support and query examples: a point graph for feature propagation and a distribution graph for uncertainty propagation. Both graphs are refined through multiple layers of message passing.

Algorithm - DPGN:

Reference: Yang, L., Li, L., Zhang, Z., Zhou, X., Zhou, E., & Liu, Y. (2020). DPGN: Distribution Propagation Graph Network for Few-Shot Learning. CVPR 2020.

Constructors

Constructor Summary
DPGNAlgorithm(DPGNOptions<T, TInput, TOutput> options) Initializes a new DPGN meta-learner.

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)