MARCH architecture enhances recurrent memory for long-context AI models

2026-08-27

Researchers have developed MARCH, a novel network architecture designed to improve the long-context retrieval capabilities of recurrent neural networks. The system addresses limitations in existing models by introducing content-routed state anchors for efficient memory scaling.

Source: arXiv · cs.LG

Reported by VERA Newswire.