We present SCI, a closed-loop, control-theoretic framework that models interpretability as a regulated state. SCI formalizes the interpretive error Delta SP and actively drives SP(t) in [0, 1] ("Surgical Precision") toward a target via a projected update on the parameters Theta under a human-gain budget. The framework operates through three coordinated components: (1) reliability-weighted, multiscale features P(t, s); (2) a knowledge-guided interpreter psi_Theta that emits traceable markers and rationales; and (3) a Lyapunov-guided controller equipped with rollback, trust-region safeguards, and a descent condition. Across biomedical (EEG/ECG/ICU), industrial (bearings/tool wear), and environmental (climate/seismic) domains, SCI reduces interpretive error by 25-42% (mean 38%, 95% confidence interval 22-43%) relative to static explainers while maintaining AUC/F1 within approximately 1-2 percentage points of baseline. SCI also reduces SP variance from 0.030 to 0.011, indicating substantially more stable explanations. Modeling interpretability as a control objective yields steadier, faster-recovering, and more trustworthy interpretive behavior across diverse signal regimes.


翻译:暂无翻译

0
下载
关闭预览

相关内容

ACM/IEEE第23届模型驱动工程语言和系统国际会议,是模型驱动软件和系统工程的首要会议系列,由ACM-SIGSOFT和IEEE-TCSE支持组织。自1998年以来,模型涵盖了建模的各个方面,从语言和方法到工具和应用程序。模特的参加者来自不同的背景,包括研究人员、学者、工程师和工业专业人士。MODELS 2019是一个论坛,参与者可以围绕建模和模型驱动的软件和系统交流前沿研究成果和创新实践经验。今年的版本将为建模社区提供进一步推进建模基础的机会,并在网络物理系统、嵌入式系统、社会技术系统、云计算、大数据、机器学习、安全、开源等新兴领域提出建模的创新应用以及可持续性。 官网链接:http://www.modelsconference.org/
Top
微信扫码咨询专知VIP会员