SimSiam<T>
Models & Types · AiDotNet.SelfSupervisedLearning
SimSiam: Exploring Simple Siamese Representation Learning.
Inheritance: ModelBase<T, Tensor<T>, Tensor<T>> → SelfSupervisedLearningMethodBase<T> → SimSiam<T>
For Beginners
SimSiam shows that simple Siamese networks can learn meaningful representations without negative pairs, momentum encoder, or large batches. The key is the stop-gradient operation applied to one branch.
How It Works
Key innovations:
Architecture:
Why it works: The stop-gradient prevents both branches from collapsing to the same constant output. The predictor makes one branch “predict” the other, creating useful gradients for learning.
Reference: Chen and He, “Exploring Simple Siamese Representation Learning” (CVPR 2021)
Best for: Simplicity, understanding non-contrastive learning.
Pros: Simplest non-contrastive method, no momentum encoder.
Cons: Can be sensitive to hyperparameters.
Constructors
| Constructor | Summary |
|---|---|
SimSiam(INeuralNetwork<T> encoder, SymmetricProjector<T> projector, SelfSupervisedLearningConfig<T> config) |
Initializes a new instance of the SimSiam class. |
Properties
| Property | Summary |
|---|---|
SelfSupervisedLearningMethodCategory Category { get; } |
|
bool RequiresMemoryBank { get; } |
|
bool UsesMomentumEncoder { get; } |
|
string Name { get; } |
Methods
| Method | Summary |
|---|---|
SimSiam<T> Create(INeuralNetwork<T> encoder, int encoderOutputDim, int projectionDim, int hiddenDim, int predictorHiddenDim) |
Creates a SimSiam instance with default configuration. |