PhD. of Alessio Lugnan
|Op fotonica gebaseerd machinaal leren om labelvrije flowcytometrie te versnellen en te vereenvoudigen
|Photonics-Based Machine Learning to Speed up and Simplify Label-Free Flow Cytometry
In this thesis, we combine arbitrary and nonlinear physical systems that require only marginal tunability and observability, using simple and computationally cheap machine learning techniques based on linear operations, such as linear regressors or classifiers. Such an approach is commonly referred to as a hardware version of reservoir computing (for dynamic hardware with feedback loops) or of extreme learning machine methods. These techniques have recently delivered state-of-the-art performance in various types of applications and they require much easier and faster training compared to their conventional counterparts in machine learning or deep learning. The role of the arbitrary nonlinear physical system is to transfer the input into a richer, higher dimensionality representation, to greatly increase the computational power of the subsequent linear machine learning model.
We apply this approach by taking advantage of the extreme processing speed of arbitrary photonic systems to classify microscopic objects, such as biological cells or microparticles, as they flow in a microfluidic channel. That is, we aim to simplify and accelerate machine learning operations in label-free microflow cytometry. Indeed, the computational costs of traditional algorithms often act as a bottleneck in this type of application, significantly limiting the throughput of online operations, such as cell sorting.
Related Research Topics
Back to list