Jamal Mohammed
PhD in Computer Science Great Distinction
Specializing in Time Series Predictability, Machine Learning for Time Series Forecasting, and Algorithmic optimization
Specializing in Time Series Predictability, Machine Learning for Time Series Forecasting, and Algorithmic optimization
Extending my PhD work on univariate time series predictability to multivariate domains, developing frameworks to quantify and estimate predictability bounds for complex, multi-dimensional temporal data.
Creating data-driven explainability and interpretability methods specifically designed for time series forecasting models, making complex temporal patterns understandable to domain experts.
Integrating my predictability estimation framework into open-source AutoML systems like Uber's Orbit, Facebook's Prophet, and Unit8's Darts to enhance automated time series analysis pipelines.
My research aims to bridge the gap between theoretical predictability bounds and practical forecasting applications. During my PhD, I developed novel methods to quantify the upper limits of predictability in univariate time series. Now I'm expanding this work to multivariate time series, which presents unique challenges due to cross-dimensional dependencies and interactions.
I'm particularly interested in how predictability estimation can enhance explainable AI for temporal data. By understanding the fundamental limits of what can be predicted in a time series, we can better evaluate model performance, identify when predictions are reliable, and explain why certain patterns are difficult to forecast.
Another key focus is integrating these concepts into open-source AutoML systems. I'm working on incorporating my predictability framework into libraries like u8darts to help data scientists automatically assess forecast difficulty and select appropriate models for their specific time series problems.
DOI:10.3390/e17042367 (First Author)
A comprehensive toolkit for implementing forecasting pipelines and evaluating prediction accuracy bounds using information theory principles.
Development of model-agnostic frameworks for estimating the upper bounds of predictability in univariate numeric time series.
Designed and implemented RNN and Bayesian neural networks for predictive maintenance and anomaly detection in industrial equipment.
My journey in computer science began at Politecnico di Torino, where I earned my Bachelor's degree in Computer Engineering. It was here that I developed a strong foundation in computational thinking and problem-solving. My passion for the field grew, leading me to pursue a Master's degree in Computer Science at the Free University of Bolzano, where I graduated with top honors (110/110, top 1%).
I recently completed my Ph.D. in Computer Science at the University of Zurich, defending my thesis with Great Distinction. My research focused on time series predictability and information theory, combining deep learning with algorithmic modeling to push the boundaries of what's possible in forecasting and predictive analytics.
My professional experience includes working as a Data Analyst at Durst Phototechnik AG, where I developed RNN and Bayesian ML models for anomaly detection. I collaborated closely with engineering teams to integrate these predictive models into maintenance pipelines, gaining valuable experience in deploying machine learning solutions in real-world scenarios.
Throughout my academic and professional journey, I've developed expertise in Python, PyTorch, TensorFlow, and various ML techniques including transformers and neural forecasting. I've also served as a Teaching Assistant for Database Systems at the University of Zurich and have had the privilege of supervising several MSc and BSc theses.
mjamal@ifi.uzh.ch
Binzmühlestrasse 14 (BIN 2 E 11), Zurich, CH-8050
University of Zurich