Memory-Augmented Network
A network bolted onto an external memory it can read from and write to, like a scratchpad it controls.
Think of it like
Like giving someone a notebook they can jot into and flip back through, instead of forcing them to hold everything in their head.
Example
Early “Neural Turing Machine” research let a network store and recall sequences using a differentiable memory bank.
How it actually works
These architectures separate computation from storage: the network learns to write facts into an addressable memory and read them back later, in principle handling tasks that need long-term recall. They were influential research but tricky to train and were largely overtaken by big transformers with long context and retrieval. The idea of external memory lives on in RAG-style systems.
For product teams
Mostly a research lineage now; its spirit survives in retrieval-based systems.
For engineers
A network with a differentiable external memory supporting learned read/write operations.
Related
- Retrieval-Augmented Architecture — The modern, practical descendant of the idea.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome