基于轨迹数据优化治疗方案，防止模型误判导致危险干预。
Conservative Continuous-Time Treatment Optimization
- 用连续时间随机微分方程建模患者动态，治疗作为连续控制变量。
- 引入路径空间MMD正则项，使治疗方案分布贴近真实观测轨迹。
- 在基准数据集上显著提升鲁棒性与性能，适合医疗决策场景。
我们提出一种保守的连续时间随机控制框架，用于从不规则采样患者轨迹中优化治疗方案。患者的未知动态被建模为以治疗为连续时间控制的受控随机微分方程。基于模型的优化若不加约束，可能利用模型误差并提出超出支持范围的控制策略，从而无法真正优化真实动态。为限制外推，我们在路径空间添加了一种基于签名的一致性MMD正则项，惩罚那些诱导轨迹分布偏离观测轨迹的治疗计划。由此得到的目标函数可计算其真实成本的上界。在基准数据集上的实验表明，该方法相比非保守基线展现出更优的鲁棒性和性能。
原文摘要 · Abstract (English)
We develop a conservative continuous-time stochastic control framework for treatment optimization from irregularly sampled patient trajectories. The unknown patient dynamics are modeled as a controlled stochastic differential equation with treatment as a continuous-time control. Naive model-based optimization can exploit model errors and propose out-of-support controls, so optimizing the estimated dynamics may not optimize the true dynamics. To limit extrapolation, we add a consistent signature-based MMD regularizer on path space that penalizes treatment plans whose induced trajectory distribution deviates from observed trajectories. The resulting objective minimizes a computable upper bound on the true cost. Experiments on benchmark datasets show improved robustness and performance compared to non-conservative baselines.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。