Browse Teach Curate

LSTM Cell State

A separate memory vector that flows along the sequence through only elementwise operations — a gated multiply and a gated add — never a full matrix transform. This additive highway is what lets information and gradients cross hundreds of steps intact.

jeremy Video

LSTM and GRU Mechanisms in Deep Learning

Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) architectures address the vanishing gradient problem in vanilla recurrent neural networks through specialized internal mechanisms known as…