Dr. Abdelkader Baggag is a Machine Learning Senior Scientist at the Qatar Computing Research Institute. He is a member of the Qatar Center for Artificial Intelligence, with a joint appointment as Associate Professor in the Information and Computing Technology Division, Hamad Bin Khalifa University, where he teaches a graduate course on Generative AI Foundations. Dr. Baggag holds a Ph.D. in Computer Science from the Department of Computer Science and Engineering, University of Minnesota, USA, with a concentration on Machine Learning and Scalable Numerical Linear Algebra.

Prior to joining the Qatar Computing Research Institute, Dr. Baggag was an academic at McGill University, and then a tenured Associate Professor at Laval University in Canada.

Dr. Baggag has extensive experience in, and in-depth knowledge of, applied machine learning and numerical methods for large-scale systems from engineering applications. He has gained this expertise while working at leading High-Performance Computing research centers, namely, the Computing Research Institute at Purdue University; the Institute for Computer Applications in Science and Engineering (ICASE) at NASA Langley Research Center in Hampton, VA; and the Army High Performance Computing Research Center, and the Minnesota Supercomputer Institute -in Minnesota, USA.

Dr. Baggag research is broad and spans all aspects of machine learning. Particular strengths are in Bayesian and numerical linear algebra approaches to modeling and inference in multimodal large language models (LLMs). The type of work ranges from studying fundamental concepts, e.g., reducing -to linear- the quadratic complexity of the Transformer, in terms of memory and computation with respect to the sequence length, all the way to getting the algorithms to perform competitively against the state-of-the-art in big-data applications.

Dr. Baggag research interests span Generative AI; Representation Learning; and multimodal Large Language Models (LLMs). Dr. Baggag is also interested in optimal transport and matrix completion -exploiting techniques about matrix functions and Random Matrix Theory for machine learning. Worked on AI and ML applications that include AI for Wearable Data Analytics, Traffic Prediction and Missing Data Imputation, and AI for Resilient Smart Cities.

Currently, Dr. Baggag is working on `Multimodal Large Language Models and Application-Driven Offline Reinforcement Learning, i.e., offline reinforcement learning research and prescriptive learning, with applications in LLMs, e.g., RLHF and implicit reward methods for alignment in LLMs such as Direct Preference Optimization (DPO).

Current projects include:

1. The Linear Algebra of Large Language Models.
2. Mass fact editing in LLM.
3. Watermarking of LLMs.

Vision -- What are the main open questions in LLMs? There are a lot of topics, e.g., going on beyond transformer architectures to learn (Mamba for example), working on small models which are highly efficient, making LLMs talk to each other (AI agents), grounding LLMs to reality with sensing capabilities, how to do large-scale distributed training. We are just starting to do the real research in LLMs. All before was based on engineering craft --which is already hard.

1. PhD in Computer Science (Summa Cum Laude), major in Machine Learning, Scalable Numerical Linear Algebra, HPC, Random Matrix Theory for Machine Learning.
2. Capacity Building in Artificial Intelligence in Qatar:
   1. Mentoring HBKU doctoral and masters students | Recruiting and training Postdocs, Scientists, Research Assistants, Research Associates and Summer Interns.
   2. A team leader experienced in guiding engineers and scientists.
   3. Expertise in Machine Learning and Artificial Intelligence, Generative AI, High-Performance Computing, Reinforcement Learning, Large Language Models.
   4. Data Analytics | Design of data-driven tools for real-world applications.
   5. Penalized models such as Lasso, ElasticNet, GroupLasso, Ridge, etc.
   6. Data representation and reduction: Nonnegative matrix factorization, data representation in domain transfer learning problem, multi-manifolds.