Recurrent Neural Networks (RNN)

Definition

A specialized neural network designed to process sequential data. Unlike standard networks, RNNs have 'memory' loops that allow information from previous steps to influence the current step.

Explain Like I'm New

Reading a book. If you read the word 'Bank', does it mean money, or a river? You can only know if you remember the sentence you just read. A normal neural network forgets everything instantly. An RNN has a 'memory loop', allowing it to remember the past context.

Real World Example

Before Transformers were invented in 2017, RNNs (specifically LSTMs) powered Google Translate and Apple's Siri, because translating language requires understanding the sequence of words.

Common Use Cases

•Time-series forecasting
•Legacy language processing
•Speech recognition

Interview Questions

basic

What type of data are RNNs primarily used for?

intermediate

Why have Transformers largely replaced RNNs in modern AI like ChatGPT?

Flash Cards

Question

Type of data?

Click to reveal answer

Answer

Sequential data (Text sentences, Audio waves, Stock market prices over time).

Question

Why replaced by Transformers?

Click to reveal answer

Answer

Because RNNs must process words one-by-one in strict order, which is incredibly slow and prevents GPUs from doing parallel processing. Furthermore, RNNs suffer from 'Short-Term Memory'—they forget the beginning of a long paragraph by the time they reach the end.