MoCo<T>

Models & Types · AiDotNet.SelfSupervisedLearning

MoCo: Momentum Contrast for Unsupervised Visual Representation Learning.

Inheritance: ModelBase<T, Tensor<T>, Tensor<T>> → SelfSupervisedLearningMethodBase<T> → MoCo<T>

Known derived types: MoCoV2<T>

For Beginners

MoCo is a contrastive learning method that uses a momentum encoder and a memory queue to provide a large pool of consistent negative samples without requiring huge batch sizes.

How It Works

Key innovations:

How MoCo works:

Reference: He et al., “Momentum Contrast for Unsupervised Visual Representation Learning” (CVPR 2020)

Best for: Limited GPU memory, consistent negative samples.

Pros: Memory efficient, consistent negative samples, good performance.

Cons: More complex than SimCLR, requires a momentum encoder.

Constructors

Constructor Summary
MoCo(INeuralNetwork<T> encoder, IMomentumEncoder<T> momentumEncoder, IProjectorHead<T> projector, IProjectorHead<T> momentumProjector, int embeddingDim, SelfSupervisedLearningConfig<T> config) Initializes a new instance of the MoCo class.

Properties

Property Summary
IMemoryBank<T> MemoryBank { get; } Gets the memory bank used for negative samples.
SelfSupervisedLearningMethodCategory Category { get; }
bool RequiresMemoryBank { get; }
bool UsesMomentumEncoder { get; }
string Name { get; }

Methods

Method Summary
MoCo<T> Create(INeuralNetwork<T> encoder, Func<INeuralNetwork<T>, INeuralNetwork<T>> createEncoderCopy, int encoderOutputDim, int projectionDim, int queueSize) Creates a MoCo instance with default configuration.
void Reset()