What is sequence-to-sequence forecasting in time series deep learning?

Sequence-to-sequence models map an input sequence to an output sequence using encoder-decoder architecture.

Seq2Seq models use an encoder to compress historical time series into a latent representation and a decoder to generate future values step-by-step. This architecture is powerful for multi-step forecasting. Attention mechanisms further improve performance by allowing the decoder to focus on relevant time steps from the input sequence.