INVITED SPEAKER 邀请报告

 

Samir Brahim Belhaouari, Hamad Bin Khalifa University, Qatar

Dr. Samir Brahim Belhaouari is a faculty member in the Division of Information & Computing Technology at Hamad Bin Khalifa University (HBKU), Qatar. He received his Ph.D. in Mathematical Sciences from EPFL, Switzerland, and his MSc in Networks and Telecommunications from INP/ENSEEIHT, France. With international academic experience across Europe, the Middle East, Russia, and Asia, he has developed a strong global research profile.
His work spans mathematics, machine learning, and data science, with notable contributions in classification, GenAI, feature selection, data preprocessing, and optimization. He is the creator of the Multilevel Architecture of Deep Learning (MADL), a framework that enhances neural network training efficiency and performance. He has also developed novel algorithms for time-frequency decomposition, model compression, Green AI, and image/graph representation, as well as an enhanced hashing function inspired by mathematical conjectures.
Dr. Belhaouari has served as principal investigator on major research projects funded by Qatar, Russian, and Malaysia, and his contributions have earned him multiple international awards, including medals at global innovation exhibitions. With over 300 publications, more than 5,000 citations, and an H-index of 36, he continues to influence both theoretical and applied aspects of AI, delivering innovative solutions that bridge mathematics, sustainability, and advanced computing.

Talk Title: Leveraging Mathematics to Address AI and Security Challenges
Abstract: Solving complex computer science problems becomes more manageable by harnessing the power of unsolved mathematical puzzles and inspiration from nature. Mathematics provides the foundation for crafting advanced algorithms in optimization, hashing, data compression, and model refinement, which are essential for tackling a wide range of challenges in artificial intelligence (AI) and cybersecurity. This talk will explore several key projects that illustrate the pivotal role of mathematics in addressing these issues.
Highlights include smart pruning techniques for deep neural networks (Green LLMs, CNNs, etc.), Green AI from Deep to shallow learning, the Walking Algorithm for Longitudinal Key Signatures (WALKS), and optimizing dimensionality reduction and visualization using enhanced clustering and optimal transport.
We will also discuss a novel time-frequency decomposition, KNNOR—a method for oversampling and downsampling imbalanced datasets—and chaos-based hashing combined with Gaussian Kernel LSH for improved data security and similarity search. Other topics include feature selection for high-dimensional data, extending the Komlós Conjecture for categorical variable encoding, and various deep learning innovations, such as architectural designs, fine-tuning methods, specialized loss functions, and activation functions. Additionally, we’ll cover anomaly detection, clustering techniques, and a dynamic Markov Chain coupled with reinforcement learning for optimization and feature selection. Applications of these techniques in biomedical and bioinformatics domains will be examined, along with the use of number theory in security, particularly in hashing and RSA encryption. Through these projects, we demonstrate how mathematical insights lead to cutting-edge solutions in AI and cybersecurity