Description
A Multiplicative LSTM (mLSTM) is a recurrent neural network architecture for sequence modelling that combines the long short-term memory (LSTM) and multiplicative recurrent neural network (mRNN) architectures. The mRNN and LSTM architectures can be combined by adding connections from the mRNN’s intermediate state to each gating units in the LSTM.
Papers Using This Method
NR4DER: Neural Re-ranking for Diversified Exercise Recommendation2025-06-01MoxE: Mixture of xLSTM Experts with Entropy-Aware Routing for Efficient Language Modeling2025-05-01Tiled Flash Linear Attention: More Efficient Linear RNN and xLSTM Kernels2025-03-18DeltaProduct: Improving State-Tracking in Linear RNNs via Householder Products2025-02-14Unlocking State-Tracking in Linear RNNs Through Negative Eigenvalues2024-11-19xLSTM: Extended Long Short-Term Memory2024-05-07Multiplicative LSTM for sequence modelling2016-09-26