Machine Learning in Aerosol Science: Something Old, Something New
Machine learning is pervading most areas of research, promising the potential to deliver new insights and solutions to topical questions relevant to science and society. Aerosol science is no different and, as a highly multidisciplinary area of research, delivers a wide range of problems to which machine learning might be applied. Indeed a recent report by the UK aerosol society notes that ‘Aerosol science is core to a broad range of disciplines extending from drug delivery to the lungs to disease transmission, combustion and energy generation, materials processing, environmental science, and the delivery of agricultural and consumer products’. This naturally provides a range of challenges when it comes to technology adaption, not least from a training and development perspective. One might argue each element of aerosol research has distinctly different drivers when it comes to such adaption and this should be managed in a siloed way. However, in some ways, the excitement and widespread use of machine learning brings its own momentum in whatever discipline it is applied. Some consider wide-scale adoption of machine learning to reflect a form of ‘solutionism’, whilst others are demonstrating a quantifiable benefit. In this talk, I will attempt to generate a debate on such issues by providing examples of adoption across aerosol science. These include, but are not limited to: new approaches for extracting information from aerosol instrumentation; methods for evaluating impacts on human health; methods for predicting fundamental properties; and techniques for reducing computational cost in process models. As already noted, aerosol science is naturally multidisciplinary and machine learning might offer a vehicle for efficient knowledge exchange if we appreciate the key factors that will dictate sustainability and wide-scale use.
For more information visit this link