AutoCorrelation Mechanism with the following two phases: (1) period-based dependencies discovery (2) time delay aggregation This bloc

Series decomposition block

Special designed layernorm for the seasonal part

Compose with two layers

Autoformer encoder

Autoformer decoder layer with the progressive decomposition architecture

Compose with two layers

Autoformer encoder

Autoformer encoder layer with the progressive decomposition architecture

Multi-Head Attention module

Two-layer position-wise feed-forward neural network.

Moving average block to highlight the trend of time series

Bottleneck convolution CSCM

A encoder model with self attention mechanism.

A encoder model with self attention mechanism.

Wrapper around SSKernel that generates A, B, C, dt according to HiPPO arguments. The SSKernel is expected to support the interface forward()

Vanilla Transformer with O(L^2) complexity

Autoformer is the first method to achieve the series-wise connection, with inherent O(LlogL) complexity

Autoformer is the first method to achieve the series-wise connection, with inherent O(LlogL) complexity

Autoformer is the first method to achieve the series-wise connection, with inherent O(LlogL) complexity

Vanilla Transformer with O(L^2) complexity

Autoformer is the first method to achieve the series-wise connection, with inherent O(LlogL) complexity

A sequence to sequence model with attention mechanism.

Stores a representation of and computes the SSKernel function K_L(A^dt, B^dt, C) corresponding to a discretized state space, where A is Normal + Low R

Scaled Dot-Product Attention