Advancing Theory and Trust in Cognitive Systems

在我们日益自动化和互联的世界中，智能系统无处不在-从智能家居到自动驾驶汽车。下一代智能系统被称为认知系统，是一种在具有感知认知量的环境中运行的系统。这些都是具有更高智能水平的智能系统，可以自主运行。从根本上讲，认知系统的安全可靠运行在很大程度上取决于其传感器、其对收集数据的理解以及其与环境的交互。基于传感器数据，认知系统可以使用机器学习技术来构建其知识库和模型。然而，这通常会创建输入-输出驱动的模型，其中对系统动态的信息或用户理解最少。这对认知系统的未来提出了一个重要的问题。我们是否可以仅仅依靠认知系统来完成关键或不安全的任务，或者应该将它们调整为被动的行动和卑微的职责？我提出了一个研究计划，在未来五年内，该计划将重点关注新颖有效的方法来推进认知系统理论，同时提高公众对其应用的信心、接受度和信任度。这些活动将支持我的研究计划的长期目标，即使认知系统更加可靠和可访问。三个主要活动将支持这一方案。开发稳健的估计和控制策略，以提高感知-动作周期内的认知系统性能。我们将比较和开发鲁棒估计器和控制器，以：1）从环境中提取尽可能多的有用信息; 2）允许可靠的数据集创建我们的系统知识库; 3）实现安全和可预测的控制动作;以及，4）提高整体系统的可靠性和性能。支持机器学习和基于物理的建模，以提高系统智能。我们将研究融合这两种模型的方法，以：1）改善决策过程的优化; 2）提高整体系统性能; 3）帮助工程师或用户更好地理解认知系统的推理或行为。为认知系统提供解释和推理其行为的能力，以提高用户信任。我们将研究和开发置信度和可解释性，以：1）用简单的语言向用户描述如何以及为什么获得特定的解决方案; 2）增加用户对认知系统的信任和接受。通过支持这些目标的活动，我的研究计划将开发更可靠和安全的认知系统，以改善人类的日常生活。

Smart systems are found everywhere in our increasingly automated and interconnected world-from smart homes to self-driving vehicles. The next-generation smart system, known as a cognitive system, is a type of system that operates in an environment with a perceived amount of cognition. These are smart systems with a higher-level of intelligence that behave autonomously. Fundamentally, the safe and reliable operation of a cognitive system is heavily dependent on its sensors, its understanding of the collected data, and its interaction with the environment. Based on the sensor data, a cognitive system may build its knowledge base and models using machine learning techniques. However, this generally creates models that are input-output driven with minimal information or user understanding on the dynamics of the system. This raises an important question on the future of cognitive systems. Can we rely solely on a cognitive system for critical or unsafe tasks, or should they be regulated to passive actions and menial duties? I propose a research program that, in the next five years, will focus on novel and efficient ways of advancing the theory of cognitive systems while improving public confidence, acceptance, and trust in their application. These activities will support the long-term goal of my research program which is making cognitive systems more reliable and accessible. Three main activities will support this program. Develop robust estimation and control strategies to improve cognitive system performance within the perception-action cycle. We will compare and develop robust estimators and controllers to: 1) extract as much useful information from the environment as possible; 2) allow for reliable datasets to create our system's knowledge base; 3) enable safe and predictable control actions; and, 4) improve overall system reliability and performance. Fuse machine learning and physics-based modelling for improved system intelligence. We will investigate methods to fuse the two types of models in an effort to: 1) improve the optimization of the decision-making process; 2) increase overall system performance; and, 3) help the engineer or user better understand the reasoning or actions of the cognitive system. Provide the ability for a cognitive system to explain and reason its actions to improve user trust. We will investigate and develop confidence factors and explainability to: 1) describe how and why a particular solution was obtained, in plain language, to the user; and, 2) increase user trust and acceptance of the cognitive system. Through the activities supporting these aims, my research program will develop more reliable and safe cognitive systems that will improve the daily lives of humans.