arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2604.25904cs.LGmath.DSstat.ML
收藏

教师强制作为广义贝叶斯:在混沌动力学中切换代理的优化几何不匹配

Teacher Forcing as Generalized Bayes: Optimization Geometry Mismatch in Switching Surrogates for Chaotic Dynamics

Interdisciplinary Center for Scientific Computing, Heidelberg University, Germany(海德堡大学交叉科学计算中心,德国);Dept. of Theoretical Neuroscience, Central Institute of Mental Health (CIMH), Mannheim, Germany(心理健康中央研究所(CIMH)理论神经科学系,曼海姆,德国);Hector Institute for AI in Psychiatry and Dept. of Psychiatry and Psychotherapy, CIMH, Mannheim, Germany(精神病人工智能研究所(Hector)及心理健康与心理治疗系,CIMH,曼海姆,德国);Hertie Institute for AI in Brain Health, University of Tübingen, Germany(脑健康人工智能研究所(Hertie),图宾根大学,德国)

Andre Herz, Daniel Durstewitz, Georgia Koppe

AI总结:

本文研究了教师强制在切换代理中的优化几何不匹配问题,通过比较教师强制与边际似然的曲率,发现单个强制路径会增加曲率,而多重切换解释会减少曲率,实验表明窗口证据微调可提升证据但可能降低动态量。

摘要:

身份教师强制(ITF)能够稳定训练确定性递归代理以模拟混沌动力系统,并在递归神经网络(RNN)中有效用于动态系统重建(DSR),包括可解释的近线性RNN(AL-RNN)。然而,作为基于干预的预测损失(和广义贝叶斯更新),教师强制不需要与自由运行模型的边际似然几何匹配。我们比较了ITF和边际似然在AL-RNN的概率切换增强中的目标诱导曲率,通过Louis的身份估计模糊意识观测信息。在本研究的切换设置中,条件于单个强制路径(如ITF所做)会增加曲率,而当多个切换解释仍合理时,边际似然曲率通过缺失信息修正而减少。在洛伦兹-63实验中,窗口证据微调提高了保留证据,但可能相对于ITF预训练模型降低了动态量。

英文摘要:

Identity teacher forcing (ITF) enables stable training of deterministic recurrent surrogates for chaotic dynamical systems and has been highly effective for dynamical systems reconstruction (DSR) with recurrent neural networks (RNNs), including interpretable almost-linear RNNs (AL-RNNs). However, as an intervention-based prediction loss (and thus a generalized Bayes update), teacher forcing need not match the free-running model's marginal likelihood geometry. We compare the objective-induced curvatures of ITF and marginal likelihood in a probabilistic switching augmentation of AL-RNNs, estimating ambiguity-aware observed information via Louis' identity. In the switching setting studied here, conditioning on a single forced regime path (as ITF does) inflates curvature, while marginal likelihood curvature is reduced by a missing-information correction when multiple switching explanations remain plausible. In Lorenz-63 experiments, windowed evidence fine-tuning improves held-out evidence but can degrade dynamical quantities of interest (QoIs) relative to ITF-pretrained models.

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。

↑