My postdoctoral Research!

  • I am involved in the following projects that apply machine learning and statistical methods to problems in real-life applications:
    • Project 1: Designed and built time series machine learning predictive models to predict fall occurrences based on thousands of high dimensional gait sequences captured from patients with dementia.
    • Project2: Designed and built supervised machine learning techniques to detect the body positions and movements made while sleeping because of their relationships to Pressure ulcers.
    • Project 3: Designed and built supervised machine learning techniques to automatically detect compensatory motions during stroke rehabilitation therapy which are common post-stroke.
    • Tools: Python, Scikit-learn, NumPy, Scipy, TensorFlow, NLTK, and XGBoost
    • ML Algorithms: Logistic regression, Deep Learning (e.g. LSTM-Recurrent Neural Network), Decision Trees, Support Vector Machines, ensemble methods (e.g. Random Forest, Gradient Boost and AdaBoost, and Gradient tree boosting) and dimensionality reduction techniques (PCA, probabilistic PCA, kernel PCA and GPLVM)
  • I am also Leading and mentoring analytic team including interns, clinical engineers and research associates at AIRR.
  • I am also documenting the findings and disseminating the results of my research through a portfolio of publications and presentations.
  • My doctoral Research!

  • In my doctoral studies, I built a significant research enterprise in healthcare technology that enhances the existing gait analysis tool through the following projects:
    • Project 1: Designed and built unsupervised machine learning model to find a pattern in high dimensional gait sequences and generate a composite measure indicative of overall gait performance. The model influenced the existing analysis in the gait clinic at a hospital in Toronto.
    • Project 2: Designed and built generative dynamic Bayesian machine learning model to automatically diagnose pathological walking condition. In this study time-series of walking pattern of several individuals were captured and analyzed. Two approaches, an instance-based discriminative classifier (k-nearest neighbor) and a dynamical generative classifier (using Gaussian Process Latent Variable Model) were used to discriminate between healthy and pathological gait patterns (>98% accuracy)
    • Project 3: Exploited exploratory data analysis and feature selection (wrapper and filter) techniques to transform raw videos of motions into analyzable features and extract the most informative and discriminatory features in the dataset
    • Project4: Used different performance measure techniques to select the final model and published results in prestigious academic journal
    • Project 5: Developed prototypes for ubiquitous sensing technology (The software prototype is available to public on GitHub)
    • Tools: Python, Scikit-learn, NumPy, Scipy, Bayes Net toolbox for MATLAB, GPML, and C++.
    • ML Algorithms: clustering (e.g. restricted Boltzmann machine, self-organizing map, model based Gaussian mixtures, RBM), generative Bayesian models such as Gaussian process, dynamic models such as Hidden Markov models and GPLVM.