MLPMixerBlockLayer<T>

Layers · AiDotNet.NeuralNetworks.Layers

A single MLP-Mixer block: temporal-axis MLP + channel-axis MLP, each wrapped with pre-norm and residual, per Tolstikhin et al. 2021 “MLP-Mixer: An all-MLP Architecture for Vision” (extended to time-series patches by Ekambaram et al. 2024 for Tiny Time Mixers).

Inheritance: LayerBase<T> → MLPMixerBlockLayer<T>

Implements: IShapeContract

How It Works

Operates on input tensors of shape [B, numPatches, hiddenDim]. The forward sequence is:

The temporal mixer mixes information across patches (time dimension); the channel mixer mixes information across hidden features at each patch position. The transpose is required so a plain DenseLayer (which operates on the last axis) can be applied across the patch axis.

Constructors

Constructor Summary
MLPMixerBlockLayer(int numPatches, int hiddenDim, int expansionFactor) Initializes a new MLPMixerBlockLayer.

Properties

Property Summary
bool SupportsTraining { get; }

Methods

Method Summary
IReadOnlyList<ILayer<T>> GetSubLayers()
IReadOnlyList<OutputAxisContract> OutputAxesFor(int inputRank) Derived from this type’s [TensorLayout] declarations.
void ResetState()
void UpdateParameters(T learningRate)