Description
The Legendre Memory Unit (LMU) is mathematically derived to orthogonalize its continuous-time history – doing so by solving d coupled ordinary differential equations (ODEs), whose phase space linearly maps onto sliding windows of time via the Legendre polynomials up to degree d-1. It is optimal for compressing temporal information.
See paper for equations (markdown isn't working).
Official github repo: https://github.com/abr/lmu
Papers Using This Method
SynthRAD2025 Grand Challenge dataset: generating synthetic CTs for radiotherapy2025-02-24Natively neuromorphic LMU architecture for encoding-free SNN-based HAR on commercial edge devices2024-07-04LMUFormer: Low Complexity Yet Powerful Spiking Model With Legendre Memory Units2024-01-20Learning-Based Data Storage [Vision] (Technical Report)2022-06-12Contrasting LMU with LSTM2022-01-17Parallelizing Legendre Memory Unit Training2021-02-22Feedforward Legendre Memory Unit2021-01-01The LMU Munich System for the WMT 2020 Unsupervised Machine Translation Shared Task2020-10-25Legendre Memory Units: Continuous-Time Representation in Recurrent Neural Networks2019-12-01