I-Corps: Analog artificial neural network (ANN) structure with tunable parameters for identification of acoustic events

I-Corps项目更广泛的影响/商业潜力是开发极低功耗系统，将机器学习应用于音频信号。 这使得连续监测复杂的环境，而无需大的电源。最初的重点是农业，在农业中，监测牲畜可能会导致更健康的动物和更低的抗生素使用。 这可能导致改善动物福利和降低动物蛋白的生产成本。此外，它可能有助于解决人类通过肉类产品消费者获得抗生素耐药性的挑战。尺寸重量和功率（SWAP）优势可能使该技术对广泛的其他领域具有吸引力，包括环境监测、制造、安全和新的便携式医疗设备。I-Corps的这一项目是基于开发调整传统人工神经网络（ANN）的方法，使其即使在使用近似计算模拟技术的情况下也能可靠工作。 在片上模拟人工神经网络的设计中使用这些方法使得在超低功耗硬件中执行音频事件分类成为可能。 此外，实用的机器学习和音频信号处理算法用于使用开发的专有稀疏矩阵编码字典算法来有效地表征环境音频。 这些算法在低功耗边缘计算硬件上运行，这些硬件在本地捕获和分析音频环境。这些算法还允许ANN的鲁棒实现，用于进一步的声学分类。当与其他涉及可编程模拟电路的解决方案相结合时，其结果将用更高效的模拟处理取代耗电的数字电路。该奖项反映了NSF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 extremely low-power systems for applying machine learning to audio signals. This enables the continuous monitoring complex environments without the need for large power supplies. The initial thrust is in agriculture where monitoring livestock may result in healthier animals and lower antibiotic usage. This may result in improved animal welfare and lower production costs for animal protein. In addition, it may help address the challenge of humans developing antibiotic resistance acquired through meat products consumer. The size-weight-and-power (SWAP) advantages may make the technology attractive to a broad range of other fields including environmental monitoring, manufacturing, security, and new portable medical devices. These innovations could result in enhancing human health and safety.This I-Corps project is based on the development of methods of adapting traditional artificial neural networks (ANNs) to work reliably even when implemented using approximate computing analog technology. Using these methods in the design of on-chip analog ANNs makes it possible to perform audio event classification in ultra-low-power hardware. In addition, practical machine learning and audio signal processing algorithms are used to efficiently characterize environmental audio using developed proprietary sparse-matrix-encoded dictionary algorithms. These algorithms run on low power edge computation hardware that capture and analyze an audio environment locally. These algorithms also allow robust implementations of ANN for further acoustic classification. When combined with other solutions involving programmable analog circuits, the result replaces power-hungry digital circuitry with more efficient analog processing.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.