Seminar M.Sc Students
Assessing Fall Risk in Parkinson's Disease and Multiple Sclerosis Using Machine Learning and Digital Mobility Outcomes from a Wearable Device: A Prospective Study
Sama Balum, M.Sc. student in the School of Industrial & Intelligent Systems Engineering
Advisors: Prof. Neta Rabin, Prof. Jeffrey M. Hausdorff
Abstract: Falls are among the most consequential and preventable complications of chronic neurological disease, affecting over 60% of individuals with Parkinson's Disease (PD) and 50–60% of those with Multiple Sclerosis (MS) annually. This study leverages the large-scale, multi-center Mobilise-D Clinical Validation Study to evaluate machine learning-based fall prediction over a one-year prospective follow-up in individuals with PD and MS cohorts. Participants wore a lower-back inertial sensor for up to seven days at baseline, from which Digital Mobility Outcomes (DMOs), including gait speed, stride variability, cadence, and turning characteristics, were derived alongside clinical assessments and self-reported measures.
This study addresses three core questions: whether DMO sensor-derived features provide meaningful and complementary predictive signal relative to clinical measures; which features are most consistently informative across disease cohorts; and whether modeling raw bout-level gait sequences with a Temporal Convolutional Network (TCN) offers advantages over traditional models trained on aggregated features.
A multi-algorithm framework (XGBoost, Logistic Regression, TCN) with subject-level cross-validation showed DMO-only models were competitive with clinical models, with combined features yielding the best performance. The TCN with self-reported features outperformed XGBoost alone, most notably in PD. SHAP and Decision Curve Analysis confirmed disease-specific feature importance and clinical utility across thresholds.
Bio: Sama Balum holds a B.Sc. in Biomedical Engineering from Tel Aviv University and is currently an M.Sc. student in the School of Industrial & Intelligent Systems Engineering, Tel Aviv University, and CMCM lab, Tel Aviv Sourasky Medical Center, supervised by Prof. Neta Rabin and Prof. Jeffrey M. Hausdorff. Her research focuses on prospective fall prediction in neurological populations using IMU-derived Digital Mobility Outcomes, interpretable machine learning, and deep learning architectures.
Low Complexity Online Contextual Learning with Continuous Actions
Mohsen Najjar, M.Sc. student in the School of Industrial & Intelligent Systems Engineering
Advisors: Dr. Ilai Bistritz
Abstract: We study an online contextual learning problem, where an agent repeatedly observes independent and identically distributed (IID) contexts $c_t \in \R^d$ and selects actions $x_t \in \R^k$ to maximize its cumulative reward $r(x_t,c_t)$ over $T$ rounds.
The reward function is Lipschitz continuous in contexts, so good actions for a given context are also reasonably good for similar contexts.
Current algorithms that leverage this structure are practically infeasible due to large runtime or memory complexity.
In this paper, we propose Congrad, a simple kernel-based projected gradient ascent algorithm, which maintains $O(n)$ memory and $O(n(k+d))$ computational complexity per iteration by projecting policies onto a fixed $n$-dimensional function space.
Congrad utilizes a kernel that at each turn updates the actions for contexts $c$ near the observed $c_t$. The kernel initially has a large bandwidth to enable fast global learning, and progressively narrows for local refinement. We prove an expected regret bound of $O(T^{\frac{d+1}{d+2}})$, independent of the action space dimension $k$.
Bio: Mohsen Najjar holds an B.Sc. in Computer Science and Mathematics from the University of Haifa. He is supervised by Dr. Ilai Bistritz at Raccoon Labs. His research focuses on online learning and stochastic optimization, with an emphasis on theoretically grounded algorithms for sequential decision-making
