Event Details
Optimizing Latent Factor Models for Recommender Systems
Presenter: Mohamed Elrfaey
Supervisor:
Date: Wed, October 1, 2025
Time: 10:30:00 - 11:15:00
Place: Zoom - see below.
ABSTRACT
Speaker: Mohamed Elrfaey, PhD Candidate, ECE
Title: Optimizing Latent Factor Models for Recommender Systems
Date/Time: Tuesday, September 30, 2025, 10:30 - 11:15 AM
Location: [Zoom:
Meeting ID: 897 0754 0450
Passcode: 558710
Abstract:
Recommender systems are at the core of modern digital platforms, driving personalized content discovery in e-commerce, entertainment, and online learning. Despite their success, these systems face persistent challenges such as data sparsity, cold-start limitations, scalability, and overfitting.
In this seminar, I will present our research on optimizing matrix factorization, a widely used collaborative filtering technique, to address these issues. Through systematic parameter tuning and the strategic injection of Gaussian noise, we demonstrate significant improvements in robustness and generalization without increasing model complexity. Our optimized model achieves a 14% gain in prediction accuracy over baseline methods while maintaining computational efficiency and interpretability.
This work highlights how simplicity, when carefully tuned, can outperform more complex alternatives. I will also discuss potential extensions, including adaptive noise scheduling, integration of auxiliary data, and broader testing across large-scale datasets such as MovieLens and Netflix.
Bio (2–3 sentences):
Mohamed Elrefaey is a PhD candidate in the Department of Electrical and Computer Engineering at the 樱花影视, supervised by Dr. T. Aaron Gulliver. His research focuses on recommender systems and personalization, with contributions spanning from optimizing latent factor models for scalability and robustness to exploring advanced multi-attention architectures for multi-modal recommendation.