Research Assistant in AI for Healthcare 02/2022 to present
Institute of Biomedical Engineering, Department of Eng. Science, University of Oxford UK
Project: InnoHK Programme 3.2: Human Intelligence and AI Integration (HIAI) for the Prediction and Intervention of CVDs: Warning System at Hong Kong Centre for Cerebro-cardiovascular Health Engineering (COCHE)
PI: Prof. David A. Clifton 12/2022 to present
Languages/Tools used: Python, Pytorch, Microsoft VS Code
Proposed and published a causal inference technique to selectively balance confounder latent factors for addressing selection-bias that enables individualised treatment effects estimation from observation data.
Proposed and published (under review) a framework for dynamic information sharing between treatment groups that enables reliable heterogeneous treatment effects estimation with limited data.
Collaborated and published work (under review) on studying the treatment effect of different drugs on risks of new-onset gastric cancer or gastric diseases in type 2 diabetes.
Collaborated with colleagues to develop and publish advanced deep learning techniques for healthcare, such as diffusion models for synthetic data generation, Graph Neural Networks for EHR representation and Natural Language Processing for cancer detection from clinical notes etc.
Project: PARADISE: Developing predictive deep learning based algorithms for post-operative atrial fibrillation in patients undergoing cardiac surgery using ICU data
Partners: UCL, Harvard Medical School, Liverpool Heart and Chest Hospital NHS Foundation Trust, Oxford University Hospital NHS Trust, Perioperative Medicine Barts Heart Centre (funded by Medical Research Council)
PI: Prof. David A. Clifton 02/2022 to 11/2022
Languages/Tools used: Python, Pytorch, Microsoft VS Code
Proposed and published a neural ordinary differential equation based Perceiver model to handle irregularity in patient data.
Developed machine learning algorithms, in collaboration with clinical partners, to predict the post-operative risk of atrial fibrillation after cardiac surgery (to be published).
Research Associate in Industrial Machine Learning 05/2019 to 02/2022
Institute for Manufacturing, Department of Eng., University of Cambridge UK
Project: Digitally Optimised Through life Engineering Services 01/20220 to 02/2022
Partners: Rolls-Royce, BAE Systems, Bombardier etc. (funded by Innovate UK & Aerospace Technology Institute)
PI: Dr. Alexandra Brintrup and Dr. Ajith K. Parlikad
Languages/Tools used: Python, C/C++, Spyder, Gurobi, Keras, Google OR-tools
Proposed, deployed and published a mathematical model for a large-scale supplier order assignment problem across the supply chain of an aerospace manufacturing company. Utilised mathematical programming and meta-heuristic algorithms to solve the problem, leading to the automation of a manual process and resulting in huge cost savings to our industrial partner.
Proposed, deployed and published (under review) a trolley optimisation problem by extending the bin packing problem to load printed circuit board (PCB) components onto trolleys and stackers to build a set of PCBs in an assembly shop. This solution led to automation of manual loading and is projected to huge money/year while also providing increased production flexibility.
Proposed and published (under review) a novel conceptual optimisation framework, named multi-tier material consolidation, by exploitation of multi-to-multi relationships in supply chain, that reduces procurement cost in a supply chain.
Proposed and published a network science approach to study the impact of nestedness on ripple effect in supply-chain networks.
Co-supervised development, deployment and publication (submitted) of machine learning algorithms to predict bid price from historical supplier contracts that helped our industrial partner to better negotiate procurement prices.
Collaboration with industrial partners to understand and solve challenging industrial problems.
In recognition of my practical contributions, I was honoured with the 2021 Institute for Manufacturing Postdoctoral Award for Research Excellence.
Project: Airline Performance & Disruption Management Across Extended Networks 05/2019 to 12/2019
Partners: Boeing, Aegean Airline, Emirates, Swiss Airline etc. (funded by Boeing)
PI: Prof. Duncan Mcfarlane and Dr. Alexandra Brintrup
Languages/Tools used: Python, Keras
Proposed and published a network science based technique to identify disruptive elements of an airline that enables airlines to build robust flight schedules.
Collaborated on analysis of root-causes of delays and their impact, and frequent delay patterns in the airline network that enabled airlines to understand their operations.
Collaboration with Boeing and different airlines to solve flight delay propagation problems.
PhD Candidate (Senior/Junior Research Fellow) 09/2015 to 08/2019
Department of Computer Science and Applications, Panjab University Chandigarh, India
Project: Solving Large-Scale Machine Learning Problems (PhD Thesis Work - funded by University Grants Commission, Govt. of India)
Supervisors: Dr. Anuj Sharma and Dr. Kalpana Dahiya
Languages/Tools used: C++, MATLAB/Octave, MEX Files
Proposed and published variance reduction techniques, viz., stochastic average adjusted gradient methods (SAAG-I, II, III and IV) using first order optimization methods to solve large-scale problems.
Proposed and published cyclic and systematic sampling techniques to reduce the overall training time of large-scale learning problems.
Proposed and published stochastic second order optimisation method, namely, STRON to deal with the large-scale learning problems that enables faster convergence as compared with first order methods.
Developed and published a C++ based library, LIBS2ML, for scalable second order machine learning algorithms.
Reviewed and published the big data challenge in machine learning, recent research directions and different areas to tackle the challenge.
Proposed and published mini-batch block coordinate optimisation framework to solve big data problems.
Published book based on PhD work, “Stochastic Optimization for Large-scale Machine Learning”, CRC Press.
Weekend Projects: Deep Learning for Indic Scripts Handwriting Recognition (online & offline)
Languages/Tools used: Python, Keras
Proposed and published CNN based architecture for faster (in a few minutes) online Gurmukhi handwriting character recognition.
Proposed and published Gurmukhi handwritten word recognition by converting online data to offline (images) and using image augmentation and transfer learning.
Proposed and published (submitted) HCR-Net, a deep learning based first script independent handwriting character recognition technique that established 26 new benchmark results across 14 languages.