arXiv · cs.AI · Alexander Bering · 7 小时前模型评分32
ZenBrain：面向自主 人工智能（AI，Artificial Intelligence） 系统的七层神经科学启发记忆架构
ZenBrain: A Neuroscience-Inspired 7-Layer Memory Architecture for Autonomous AI Systems
导读
ZenBrain 提出七层神经科学启发的大语言模型智能体记忆架构，开源库版本 0.5.1 中衰减仅作用于工作记忆且无重放。在 LongMemEval_S 的 500 道题上，被评测分支二分类判定正确率为 47.7%，Letta 与 Mem0 分别为 48.4% 和 47.2%；该分支以全上下文基线 1/109.7 的输入 token 达到其 91.3% 的准确率。
分类置信度
中 · 待人工复核
- 只有单一媒体来源，需要人工复核。
来源级别：媒体与个人
事件类型：研究发布
证据摘录
- “ZenBrain is a seven-layer, neuroscience-derived architecture for agent memory in LLM agents.”
- “In the open library at version 0.5.1, decay acts only on working memory and nothing is replayed.”
- “The ablation runs in a simulation harness that switches fifteen components as stand-ins at simulated values, not as production code.”
- “On LongMemEval_S (500 questions), the evaluated ZenBrain arm - a Python port of a scoring class from our test harness, not the seven-layer system - answers 47.7% of the questions correctly under a binary judge (95% interval [43.3, 52.1]).”
- “On LoCoMo, the routing results of version 8 came from placeholder embeddings; with text-embedding-3-small, the flat variant (F1 0.059) lies above the full system (0.058).”
来源：arXiv · cs.AI · arxiv.org