Develop and test patient and disease representations for clinical models. Jointly learning embedding for different domains and further fine tuning or enhancing embeddings for some domains when limited amounts of data is available present significant AI challenges that will force us to produce methodological innovations. We build a comprehensive library for representation learning methods and develop novel methods to address these presented issues. These representation algorithms will be assessed based on their sensitivity to data characteristics including fairness. Role I technically lead the task force for patient representation for developing novel fair AI models. IBM Research. Oct 2020 - Present .
Develop novel computational disease models that could be used to identify factors that impact the rapid/slow progression of Type 1 Diabetes ("T1D") in infants and young children who are genetically pre-disposed and may be pre-symptomatic clinically. Such methodologies and models may (1) incorporate heterogeneous features coming from multiple sites and assessments covering multiple aspects of T1D. (2) leverage diverse data sets to accommodate noise and uncertainty in study data. (3) provide comprehensive view of risk factors that impact the onset of T1D in different time horizons. Role I lead the efforts of developing AI models for disease progression modeling, which lead to publications in high-impact journals such as Lancet D&E (IF 45), Lancet C&A (IF 38), Diabetes Care (IF 19), etc. IBM Research. May 2017 - May 2022 .
This work is focused on identifying correlates of individual pain, defining objective pain metrics, and determining changes to the device parameters that provide the user with the greatest impression of relief. A number of data streams will be studied to identify pain correlates from physiological signals, environmental stresses, and individual behavioral responses. These correlates will be used to identify pain metrics, understand the individual patient's pain experience and how the device parameters can be changed to optimize the sense of change in a manner that is adaptive to the varying needs of the patient. Role I am the technical leader for developing AI algorithms to define and build an objective pain metric and identifies pain correlates. IBM Research. Oct 2016 - Oct 2019.