Researches on Whispering Gallery Microcavities and Machine Learning

拟议工作的性质是开发一个研究计划，其中两个看似无关的领域，回音壁微腔和机器学习，是连接在一起的。在技术光谱的一侧，回音壁微腔是能够进行单分子检测和非线性光学驱动的微米尺寸的光学腔。另一方面，机器学习算法对于发现从高度非线性系统生成的高维数据中的数据模式是非常宝贵的，尽管它们的高功耗和对计算资源的大量需求。我们提出的跨学科研究计划包含三个相互关联的研究子计划：a）开发频率微梳并将其用于超高分辨率单分子精密光谱学，B）使用机器学习算法进行单分子光谱分析，以及c）开发用于通用机器学习任务的紧凑，快速和节能的学习机器。在子程序a）中，我们将开发一种使用锂离子酸盐回音壁微腔的频率微梳，并在水环境中操作。这种微梳将浸入含有蛋白质分子的液体悬浮液中。当单个分子落在微梳表面时，梳线的光谱变化将被记录下来。通过电光调谐，将实现超高光谱分辨率。 在子程序B）中，利用机器学习的能力，我们将分析从子程序a）获得的光谱。我们将使用深度学习技术，如前馈，卷积和递归神经网络，自动编码器，生成对抗网络等，识别分子，识别其物理和化学组成，结构折叠以及理解和预测单个分子之间的相互作用。 在子程序c）中，我们将利用锂离子电池和频率微梳的独特性质来开发微米级集成光子学习机器，这些机器可以快速但节能的方式计算各种机器学习模型。 这项研究将为单分子的基础研究带来重要价值，这是早期癌症诊断和药物工程等许多领域的关键因素。我们的研究对环境保护也很重要，因为它可以监测采矿场或石油生产废水中超低浓度的化学成分。考虑到目前普遍使用的机器学习，这种新型的光子学习机器将节省大量的能源。其无与伦比的快速处理速度在即将到来的5G和物联网革命中至关重要，在自动驾驶等领域需要超低延迟的机器学习。通过该计划培训HQP将为相关行业带来高素质的人才，从而为加拿大经济做出重大贡献。

The nature of the proposed work is to develop a research program where two seemingly unrelated fields, whispering gallery microcavities and machine learning, are connected. On one side of the technology spectrum, whispering gallery microcavities are micron sized optical cavities capable of single molecule detection and nonlinear optical actuation. On the other side, machine learning algorithms are invaluable for discovering data patterns in high dimensional data generated from highly nonlinear systems despite their high power consumption and large demand on computation resources. Our proposed interdisciplinary research program contains three inter-related research sub-programs: a) develop a frequency microcomb and use it for ultra high resolution single molecule precision spectroscopy, b) single molecule spectral analysis using machine learning algorithms and c) develop a compact, fast and energy efficient learning machine for general purpose machine learning tasks. In sub-program a), we will develop a frequency microcomb using a lithium niobate whispering gallery microcavity and operate it in aqueous environments. This microcomb will be immersed in liquid suspension with protein molecules. When a single molecule lands on the microcomb surface, spectral change of the comb line will be recorded. Through electro-optical tuning, ultra-high spectral resolution will be achieved. In sub-program b), utilizing the power of machine learning, we will analyze the spectra obtained from sub-program a). We will use deep learning technologies such as feedforward, convolutional and recurrent neural networks, autoencoder, generative adversarial networks, etc., to identify molecules, recognize their physical and chemical composition, structural folding as well as understand and predict the interactions between single molecules. In sub-program c), we will utilize the unique properties of lithium niobate and the frequency microcomb to develop micron sized integrated photonic learning machines that can compute various machine learning models in a fast but energy efficient way. The proposed research will bring significant value to fundamental research on single molecules, which is a key element in many areas such as early cancer diagnosis and drug engineering. Our research is also important to environmental protection as it enables the monitoring of ultra low concentrations of chemical components in waste water bodies from mining sites or oil production. The novel photonic learning machine will result in enormous energy savings considering the currently prevalent usage of machine learning. Its unbeatable fast processing speed is critical in the upcoming 5G and internet of things revolution where ultra low lantency of machine learning is required in areas such as autonomous driving. The training of HQP through this program will bring to the related industry highly qualified personnels and thus contribute significantly to the Canadian economy.