Frequency domain nonlinear optical conversion for efficient high-dimensional quantum processing

工业界和学术界正在努力创造新的“智能”服务，以实现自动驾驶汽车、计算机视觉和医疗诊断等应用。这些服务依赖于机器学习概念来创建可以显示类似于人类的智能行为的设备。水库计算（RC）是一种新方法，其工作方式与笔记本电脑/智能手机中的电子处理器有着根本不同。利用系统动力学，可以训练RC识别并响应不同的输入信号给出特定的输出。这使它们能够充当模拟计算机，可以执行信号分类，时间序列预测和音频/视觉识别等任务。本项目的目标是开发一种新型的紧凑和节能的光学RC平台，利用微谐振器频率梳的非线性动力学，并能够以光速处理数据。我们的研究团队将与我们的三个工业合作伙伴（TeraXion，'E' Machine Learning和Novacam Technologies）一起：（i）利用RC进行高速数据恢复和无噪声放大。此外，我们将（ii）应用RC处理在一个廉价的光探测和测距（LIDAR）系统。最后，我们将（iii）将我们的平台应用于实时光学相干断层扫描（OCT），这是一种计算要求很高的成像技术，用于例如，视网膜轮廓和手术指导该项目将有助于实现将机器学习方法纳入光学信号处理系统的长期愿景。这反过来将允许使用光子芯片以前所未有的数据速率（~1 THz）进行超快计算，直接与智能传感器和新兴的物联网兼容。该项目还将通过培训具有光学研究、多媒体处理和机器学习等多学科专业知识的高素质人才，使加拿大劳动力受益。我们技术的商业化将促进加拿大工业的增长，包括电信、卫生和人机互动等众多高科技领域。

Industry and academia are working on creating new 'smart' services to enable applications such as self-driving cars, computer vision and medical diagnosis. These services depend on machine learning concepts for creating devices that can display intelligent behavior similar to humans. Reservoir comput(ing/ers) (RC) is a new approach that works in a fundamentally different way from electronic processors in laptops/smartphones. Making use of system dynamics, RCs can be trained to recognize and give specific outputs in response to different input signals. This allows them to act as analog computers that can perform tasks such as signal classification, time-series prediction and audio/visual recognition. The goal of this project is to develop a novel type of compact and energy-efficient optical RC platform, utilizing the nonlinear dynamics of microresonator frequency combs, and capable of processing data at the speed of light. Our research team will, together with our three industrial partners (TeraXion, 'E' Machine Learning and Novacam Technologies): (i) exploit RC for high-speed data restoration and noiseless amplification. Additionally, we will (ii) apply RC for processing in an inexpensive light detection and ranging (LIDAR) system. Lastly, we will (iii) apply our platform to real-time optical coherence tomography (OCT), which is a computationally-demanding imaging technique used in, e.g., retinal profiling and surgical guidance. The project will contribute to the long-term vision of incorporating machine-learning methods into optical signal processing systems. This in turn will allow ultrafast computation at unprecedented data rates (~1 THz) using photonic chips, directly compatible with smart sensors and the emerging Internet of Things. The project will also benefit the Canadian workforce, via the training of highly-qualified personnel with multidisciplinary expertise in optics research, multimedia processing and machine learning. The commercialization of our technology will promote the growth of Canadian industries, across numerous high-tech sectors such as telecommunications, health and human-machine interaction.