State Space Model
State space models (SSMs) are a powerful class of models used to represent dynamic systems by tracking their hidden states over time. Current research focuses on developing efficient SSM architectures, such as Mamba and its variants, to overcome limitations of traditional methods in handling long sequences and high-dimensional data, particularly in applications involving time series forecasting, image processing, and dynamic system modeling. These advancements are improving the accuracy and scalability of SSMs across diverse fields, leading to significant improvements in areas like medical image analysis, autonomous driving, and natural language processing. The resulting models offer a compelling alternative to computationally expensive methods like transformers, while maintaining or exceeding performance in many applications.
Papers
From pixels to planning: scale-free active inference
Karl Friston, Conor Heins, Tim Verbelen, Lancelot Da Costa, Tommaso Salvatori, Dimitrije Markovic, Alexander Tschantz, Magnus Koudahl, Christopher Buckley, Thomas Parr
Long Range Switching Time Series Prediction via State Space Model
Jiaming Zhang, Yang Ding, Yunfeng Gao