I-Corps: Embedded Artificial Intelligence and Machine Learning for Drone Vision

这个i-Corps项目的更广泛的影响/商业潜力是，它将使人工智能和机器学习(AI/ML)解决方案能够在嵌入式和远程设备中实施。它提供了使任何相机、遥控器或探测器真正智能化并能够识别和预测的手段；任何已经配备了电子设备的设备都将具有利用决策智能的潜力。它将成为催化剂，使AI/ML真正无处不在，成为人类体验的商品。智能传感器将有助于使世界更加安全，摄像机将能够在悲剧发生前发现异常并向当局发出警报，探测器将能够预测紧急情况，而不是简单地遥测数据。在经济上，通过人类的帮助，嵌入式AI/ML将使社会更有效率，产生更大的收益。这个i-Corps项目进一步开发了人工智能和机器学习(AI/ML)的直接硬件电路的实现。AI/ML已经彻底改变了几个技术行业，如语音识别、图像处理、医疗保健和个性化广告。AI/ML通过训练数学模型(通常是非线性函数的组合)来对模型的环境做出推断。然而，AI/ML主要局限于拥有大量计算资源的基于云的应用。相比之下，这项技术将AI/ML模型实现为适合于单个集成电路的硬件电路。该项目建立在以前的技术成果的基础上，通过软件辅助编译在硬件设计中实现了软件设计原则，如可扩展性、面向服务的体系结构、模块化和松散耦合。创建软件辅助以自动重组硬件设计使硬件能够根据一系列参数更动态地变化。通过将类似的技术应用于AI/ML，该技术将能够解构和分析AI/ML模型，并通过一系列算法构建匹配的硬件电路。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势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 that it will enable artificial intelligence and machine learning (AI/ML) solutions to be implemented in embedded and remote devices. It provides the means to make any camera, remote sensor, or detector truly intelligent and capable of discernment and prediction; any device that has already been electronically equipped will have the potential to utilize decision making intellect. It will be a catalyst in making AI/ML truly ubiquitous and a commodity of the human experience. Intelligent sensors will help make the world safer, cameras will be able to detect anomalies and alert authorities before tragedies, and detectors will be able to predict emergencies instead of simply telemetering data. Economically, through human assistance, embedded AI/ML will enable society to be more efficient and produce greater yield. This has the potential to be a transformative technology that impacts multiple levels of society.This I-Corps project further develops an implementation of artificial intelligence and machine learning (AI/ML) directly as a hardware circuit. AI/ML have revolutionized several technical industries such as voice recognition, image processing, healthcare, and personalized advertisements. AI/ML operates by training a mathematical model, often a combination of non-linear functions, to make inferences about the model's environment. However, AI/ML has been primarily restricted to cloud-based applications with large quantities of computing resources. In contrast, this technology implements the AI/ML model as hardware circuit that fits within a single integrated circuit. This project builds off previous technical accomplishments that implemented software design principles, such as extensibility, service-oriented architecture, modularity, and loose coupling, in hardware design through software assisted compilation. Creating a software aid to automatically restructure the hardware design enabled the hardware to change much more dynamically according to a series of parameters. By applying similar techniques to AI/ML, the technology will be able deconstruct and analyze the AI/ML model, and through a series of algorithms construct a matching hardware circuit.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.