AI Dictionary › Modelli AI
The GRU is a type of recurrent unit for sequence processing that regulates the flow of information across time steps through gating mechanisms, meaning gates that decide how much past information to keep and how much new information to incorporate. It was proposed as a simpler alternative to LSTM units, retaining much of their ability to handle long-range dependencies in sequences.
Unlike the LSTM, which uses three distinct gates and a memory state separate from the hidden state, the GRU uses two gates, called the update gate and the reset gate, and a single state that serves as both memory and output at each time step. The update gate decides how much of the previous state to keep, while the reset gate decides how much of the past information to ignore when computing the new candidate content. This leaner structure reduces the number of parameters compared to the LSTM, for the same hidden state size.
It was widely used in natural language processing and time series analysis tasks before the spread of architectures based exclusively on attention, and it remains a relevant choice in contexts with limited computational resources or where the inherently sequential nature of the problem makes recurrent networks a natural fit.
The name refers directly to the "gates" that regulate the flow of information within the recurrent unit, and it was proposed in the first half of the 2010s as part of research into recurrent architectures for machine translation and other sequential tasks.
From our network
Kaimaki Web: Websites That Win Customers
Custom websites, web apps and digital marketing for growing businesses.
Visit kaimakiweb.com →From the Agora Intelligence blog
📱 Download the Android app (beta) iOS coming soon
Say what you mean. Get what you need.
Grace Certified, the AI coach that trains and certifies your prompt engineering, by Agora Intelligence.