SCM: Sleep-Consolidated Memory with Algorithmic Forgetting for Large Language Models

Saish Shinde · 2026

Read SCM: Sleep-Consolidated Memory with Algorithmic Forgetting for Large Language Models by Saish Shinde (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-20943.txt.

Description

Shinde presents SCM, a memory design for large language models modeled on sleep consolidation in the brain. It combines a small working memory, importance tagging, offline consolidation with distinct sleep-like phases, deliberate value-based forgetting, and a self-model for introspection. Across eight benchmark tests the prototype recalls perfectly over ten-turn conversations while cutting memory noise by 90.9 percent through adaptive forgetting.

Catalog record

CREATORSaish Shinde
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE37629
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.20943

Browse the full library · About the Tower of Babel