I-Corps: Fast and Accurate Artificial Intelligence/Machine Learning Solutions to Inverse and Imaging Problems

I-Corps项目更广泛的影响/商业潜力是开发机器学习软件，为具有量化不确定性的逆成像问题提供解决方案。所提出的技术可用于发现石油和天然气储层的位置，发现地热热点和地震断层的位置以收集清洁地热能，发现关键和稀有矿藏的位置以进行开采，以及人体内部成像以进行医学诊断。目标是通过使用拟议的软件平台来降低运营成本；例如，石油公司（或为石油公司提供服务的公司）可以部署该软件来实时监控、控制和做出关键决策。软件的预测和不确定性量化（UQ）能力可能有助于决策制定。I-Corps项目基于两种先进算法的发展：(1)模型约束机器学习方法，用于获得具有量化不确定性的实时逆/推理解；(2)模型约束方法，用于获得具有量化不确定性的实时正解。提出的技术汇集了随机规划，概率论，应用数学和机器学习的进步。所提出的软件旨在提供正向和反向功能。在任何一种模式下，它都可以使用训练数据和控制方程来产生代理模型，即训练有素的深度神经网络，既尊重数据又尊重控制方程。代理模型可以实时地为新情景提供准确的预测和相关的不确定性。此外，训练数据也可以根据用户需求进行定制。例如，石油公司（或为石油公司提供服务的公司）可以部署拟议的软件来实时监控、控制和做出关键决策。这些功能减少了所需的人力、时间和资金，同时消除了开发和维护成本，并降低了更新和操作成本。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The broader impact/commercial potential of this I-Corps project is the development of machine learning-enabled software to provide solutions to inverse and imaging problems with quantified uncertainty. The proposed technology may have application in the discovery of the location of oil and gas reservoirs, the discovery of the location of geothermal hotspots and seismic faults for harvesting clean geothermal energy, the discovery of the location of critical and rare mineral deposits for extraction, and the imaging of the interior of the human body for medical diagnosis. The goal is to reduce operational cost through the use of the proposed software platform; e.g., oil companies (or companies providing services to oil companies) may deploy this software to monitor, control, and make critical decisions in real time. The prediction and uncertainty quantification (UQ) capabilities of the software potentially may assist in decision making.This I-Corps project is based on the development of two advanced algorithms: (1) model-constrained machine learning methods to obtain real-time inverse/inference solutions with quantified uncertainty and (2) model-constrained methods to obtain real-time forward solutions with quantified uncertainty. The proposed technology brings together advances from stochastic programming, probability theory, applied mathematics, and machine learning. The proposed software is designed to provide both forward and inverse capabilities. In either mode, it can use training data and the governing equations to produce surrogate models, i.e., trained deep neural networks that respect both data and governing equations. The surrogate models may provide accurate predictions and associated uncertainties in real time for new scenarios. In addition, the training data may be customized according to the user needs. For example, oil companies (or companies providing services to oil companies) may deploy the proposed software to monitor, control, and make critical decisions in real time. These capabilities reduce the required human power, time, and finances, while eliminating development and maintenance costs and reducing update and operational costs.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.