发现信息
- 作者:Syed Muntasir Mamun
- 年份:2026
- 发表源:Zenodo (CERN European Organization for Nuclear Research)
- OpenAlex ID:https://openalex.org/W7214541402
- DOI:https://doi.org/10.5281/zenodo.22993103
- arXiv:待核验
- 检索主题:self-evolving
- OpenAlex 引用数快照:0(2026-11-12)
摘要线索
The convergence of self-evolution, dynamic memory adaptation, sparsity optimization, and information-geometric principles represents a transformative paradigm in the development of artificial intelligence. Historically, machine learning has relied upon static foundation models trained on massive, immutable datasets, resulting in inference engines that remain structurally frozen during deployment. This article synthesizes recent advancements across the domains of Large Language Models (LLMs) and Reinforcement Learning (RL) systems, proposing a unified, multi-layered framework for autonomous agents capable of evolving their cognitive and architectural structures at test-time. Drawing from meta-evolutionary memory frameworks such as MemEvolve and Evo-Memory, dual-axis sparsity mechanisms like Engram and Nemotron 3, and geometrically robust optimization techniques including Score-Matched Actor-Critics (SMAC), this report examines how these disparate components algorithmically synergize to enable robust, data-efficient intelligence. Furthermore, the reliance on human-annotated data is challenged by the integration of data-free Mamun, SM (2026) …………………………………………………1 Self-Evolving AI Agents: A Unified Framework Integrating Memory Dynamics, Sparsity Mechanisms, and Geometric Robustness in Large Language Models and Reinforcement Learning Systems evolutionary loops, exemplified by the Dr. Zero framework, which bootstraps agentic reasoning through automated, search-augmented curricula. Also available on Academia.edu: https://www.academia.edu/attachments/131877570/download_file?s=portfolio
核验与筛选
- 当前状态:待筛选、去重和一手来源核验。
- OpenAlex 元数据和引用数只用于发现,不作为正式引文或质量判断。
- 记录正式版本、venue、方法细节、实验协议和局限后,才迁入稳定论文卡片。