The research of Hojjat Salehinejad, Ph.D., spans several topics in machine learning: optimization and signal processing; graph-representation learning; representation learning from limited-imbalanced data; efficient machine learning on resource-limited devices; internet of things, wireless sensing and detection; and human-machine interaction.
Dr. Salehinejad also focuses on applied artificial intelligence (AI) in health care, particularly multimodal patient representation and learning. Explainable AI in health care approaches aim to explore and discover underlying patterns in patient data and their relationship with clinical outcomes. The ultimate objective is to develop and implement novel generalized AI solutions as decision-making support tools in practice to improve patient outcomes.
- Graph-representation learning. Investigating the design and generalization performance enhancement of graph neural networks and graph signal processing models for unimodal and multimodal unstructured patient representation learning.
- Learning from limited-imbalanced data. Developing methods for learning from limited-imbalanced clinical spatial and spatial-temporal data, including polar transformations for deep networks and synthesizing pathology in chest X-rays with generative models.
- Human-machine interaction. Utilizing wireless signals, imaging and AI to develop novel privacy-preserving, cross-domain and multiuser human activity recognition technologies. These technologies can improve the current state of remote monitoring and enhance the experiences of patients and staff when interacting with devices in smart environments.
- Efficient deep learning. Building resource-efficient deep learning models for deployment in practice, targeting privacy-preserving learning and inference on remote devices. These contributions can decrease inference time, reduce overfitting, and increase the mobility of deep learning models for federated learning and edge computing applications.
Significance to patient care
Deploying novel early warning systems in practice may reduce wait times, help identify patients with critical conditions and prioritize early intervention when necessary. Care quality also can be improved by implementing novel tools to increase diagnostic and treatment decision-making confidence, as well as foreseeing complexities and side effects prior to intervention. These systems can reduce patients' recovery time and length of stay from intervention to discharge and readmission rates — ultimately targeting improving outcomes.
- Senior member, Institute of Electrical and Electronics Engineers (IEEE), 2022-present.
- Special session co-organizer, "Signal/Image Processing-Machine Learning for Physiological Signals," 24th International Conference on Digital Signal Processing, 2023.
- Special session co-organizer, "Smart health care delivery through advancing health information technology," 19th International Conference on Automation Science and Engineering, IEEE, 2023.
- Co-chair – Publication, International Symposium on Personal, Indoor and Mobile Radio Communications, IEEE, 2023.
- Guest editor, Special Issue on "Machine Learning for Human Activity Recognition,"
Journal of Imaging, 2022-2023.
- Program committee member, Learning on Graphs Conference, 2022.
- Natural Sciences and Engineering Research Council (NSERC) of Canada Postdoctoral Fellowship in Artificial Intelligence, Evaluated as Outstanding by the Selection Committee for Computer Science, 2022.
- Doctoral Completion Award, University of Toronto, 2021.
- Edward S. Rogers Sr. Graduate Scholarship, University of Toronto, 2016.
- Ontario Trillium Scholarship, Ontario Government and University of Ontario Institute of Technology, 2016.