Browse Teach Curate

State Space Models for Sequence Modeling in Deep Learning

Sequence models that maintain a linear recurrent hidden state governed by learned state-transition, input, and output matrices, giving inference cost that is linear rather than quadratic in sequence length. This family (including structured variants such as S4 and Mamba) is an alternative to attention for very long sequences.

jeremy Video

Mamba Selective State Space Models explained in Machine Learning

Explains structured state-space models and Mamba: the linear state-update view, the equivalent convolutional form, selectivity via input-dependent parameters, and why SSMs scale linearly in sequence …