Table 7 Complexity of different model architectures.
Method | Encoder | Decoder |
|---|---|---|
Transformer | O(T)2 | \(O\left( {H\left( {H + T} \right)} \right)\) |
Autoformer | \(O\left( {TlogT} \right)\) | \(O\left( {\left( {\frac{T}{2} + H} \right)} \right.\left. {\log \left( {\frac{T}{2} + H} \right)} \right)\) |
Informer | \(O\left( {TlogT} \right)\) | \(O\left( {H\left( {H + logT} \right)} \right)\) |
Fedformer | \(O\left( T \right)\) | \(O\left( {\frac{T}{2} + H} \right)\) |
Ours | \(O\left( {\frac{{DT^{2} }}{{S^{2} }}} \right)\) | \(O\left( {\frac{{DH\left( {H + T} \right)}}{{S^{2} }}} \right)\) |
