arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2602.09075cs.LGcs.AI
收藏

在基于注意力的模型中学习记忆、学习和遗忘

Learning to Remember, Learn, and Forget in Attention-Based Models

University of Cambridge(剑桥大学)

Djohan Bonnet, Jamie Lohoff, Jan Finkbeiner, Elidona Shiqerukaj, Emre Neftci

AI总结:

提出Palimpsa模型,将上下文学习视为持续学习问题,通过贝叶斯元可塑性解决稳定性-可塑性困境,显著提升记忆容量,在MQAR和常识推理任务上优于基线。

摘要:

Transformer中的上下文学习(ICL)作为一种在线联想记忆,被认为是其在复杂序列处理任务中高性能的基础。然而,在门控线性注意力模型中,这种记忆具有固定容量且容易受到干扰,尤其是对于长序列。我们提出Palimpsa,一种自注意力模型,将ICL视为必须解决稳定性-可塑性困境的持续学习问题。Palimpsa使用贝叶斯元可塑性,其中每个注意力状态的可塑性绑定到一个由捕获累积知识的先验分布支撑的重要性状态。我们证明各种门控线性注意力模型作为特定的架构选择和后验近似出现,并且Mamba2是Palimpsa的一个特例,其中遗忘占主导。这一理论联系使得任何非元可塑性模型都能转化为元可塑性模型,从而显著扩展其记忆容量。我们的实验表明,Palimpsa在多查询联想回忆(MQAR)基准和常识推理任务上始终优于基线。

英文摘要:

In-Context Learning (ICL) in transformers acts as an online associative memory and is believed to underpin their high performance on complex sequence processing tasks. However, in gated linear attention models, this memory has a fixed capacity and is prone to interference, especially for long sequences. We propose Palimpsa, a self-attention model that views ICL as a continual learning problem that must address a stability-plasticity dilemma. Palimpsa uses Bayesian metaplasticity, where the plasticity of each attention state is tied to an importance state grounded by a prior distribution that captures accumulated knowledge. We demonstrate that various gated linear attention models emerge as specific architecture choices and posterior approximations, and that Mamba2 is a special case of Palimpsa where forgetting dominates. This theoretical link enables the transformation of any non-metaplastic model into a metaplastic one, significantly expanding its memory capacity. Our experiments show that Palimpsa consistently outperforms baselines on the Multi-Query Associative Recall (MQAR) benchmark and on Commonsense Reasoning tasks.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑