集成 Transformer 的自学习智能体
人工智能
2025-04-04 v1
摘要
本文介绍了一种自学习智能体，将 LLaMA 3.2 与渐进神经网络 (Progressive Neural Network, PNN) 集成，用于对话式 AI 和代码生成中的持续学习。该框架动态收集数据，利用最小样本进行任务微调，并通过 Meta-Learning 实现快速适应。LoRA 优化微调，弹性权重Consolidation (EWC) 提升知识保留。实验结果表明，方法在适应性和记忆稳定性方面均取得改进，为人工通用智能 (Artificial General Intelligence, AGI) 的可扩展步骤提供了潜力。
关键词
引用
@article{arxiv.2504.02489,
title = {The Self-Learning Agent with a Progressive Neural Network Integrated Transformer},
author = {Ajay Sivakumar and Shalini and Vasantha Raj and Sebastian Sylvester},
journal= {arXiv preprint arXiv:2504.02489},
year = {2025}
}
备注
7 pages, 2 figures, focuses on continual learning with PNN and LLaMA. Experiments demonstrate scalability and lifelong learning capabilities