Journal of Data Mining in Genomics & Proteomics

Lifelong machine learning in Digital Twins

9^th Global Summit on Artificial Intelligence and Neural Networks

September 07, 2021 | Webinar

S B Mahalakshmi

GE Research, Bangalore, India

Scientific Tracks Abstracts: J Data Mining Genomics Proteomic

Abstract:

Digital twin is a digital replica of a living or a non-living entity, they are simple but detailed.When an entity or a physical asset experiences a different environment, the digital twin recalibrates itself to reflect its physical counterpart. While a digital twin represents a process or product in some of the key industries like aerospace and power, in healthcare, it can represent an individual patient. GE has used digital twins in the area of power, aerospace and renewables. 93% increased reliability and 40% reduction in increased maintenance on an average for all monitored assets in less than a year are some of the outcomes enabled by the Digital Twins. While there are many facets of the role an AI can play in a digital twin, the most challenging is the ability of the twin to continously learn and stay representative of its physical counterpart even in changing environments. This means the digital twin should have the ability to continously learn to solve new problems in its domain and not unlearn the things it has already learnt. Take the case of defect detection in industrial asset, to realise the advantage of a digital twin we need to go from learning just the defect classes during training to learning novel classes of defects in a continoulsy learning paradigam. In patient monitoring the data would need to continously come from the lived experience, genetics and variability of a disease in pateints over time and use that to learn new patterns in the disease progression. Hence, these knowledge representations not only need to accept vast, multidimensional, structured or unstructured data, they will need to translate into clear outcomes continuously learning to adapt from new data.

Biography :

S B Mahalakshmi has completed her Masters in Computer Applications from Madurai Kamaraj university, 2001 and is working as a Senior Machine Learning Scientist at GE Research. Her expertise lies in building advanced sensors and digital twins that uses physics driven AI predictive models for the Aviation, Transportation and Healthcare including Lifesciences sectors. Some key areas include developing artificial intelligence methods for prognostics of battery life, generator and drive train health analytics, signal segementation for fault detection and eddy current image processing for Non-destructive testing (NDE), prediction of cell growth in Industrial Bioreactors. She has 9 patents and 9 publications in various National and International conferences.