Work

paper2-hcma-acm

Year: 2026

Type: preprint

Abstract: HCMA-ACM (paper 2): integrates HCMA (Hippocampal-Cortical Memory Architecture) as the governed long-term memory of an agentic context management (ACM) agent. ACM losslessly offloads context but preserves superseded versions of facts; HCMA consolidates them through bi-temporality, negative voting, and version collapse, so retrieval returns only the current version. On a controlled benchmark of superseding API specifications, HCMA reduces stale-value use from 100% to 0% (Llama-3.1-8B) and from 20% to 0% (GPT-4o-mini), lifting task success from 0% to 100% and 80% to 100%, respectively.

Source: OSF Preprints (OSF Preprints)

DOI: 10.17605/osf.io/rdc29

Author: Mauro Javier Silva

Language: English

Topic: AI-based Problem Solving and Planning

Subfield: Artificial Intelligence

Field: Computer Science

Domain: Physical Sciences

Open Access status: green

OpenAlex ID: W7224541397

Cites:

Cited by:

Related to: 10