Abbas Saliimi Lokman

Ts. Dr. Abbas Saliimi Bin Lokman is a Malaysian computer scientist and professional technologist who serves as a lecturer at the Faculty of Computing, Universiti Malaysia Pahang Al-Sultan Abdullah (UMPSA). A specialist in multimedia computing and artificial intelligence, his research primarily focuses on chatbots, natural language processing, retrieval-augmented generation (RAG) systems, and industrial AI applications such as predictive maintenance. He holds a Ph.D. in computing, actively leads industrial grants, and is a key member of the university's Virtual Simulation & Computing (ViSiC) research group.

Keynote Title: How LLMs Work: Understanding the Transformer Architecture Behind Frontier AI Models and Applications

Keynote Abstract: 

Large Language Models (LLMs) have transformed the way people interact with computers by enabling machines to understand the context of natural human language and generate human-like responses across multiple modalities, including text, images, video, and audio. At the core of these models is the transformer architecture, which introduced a fundamentally new approach to processing sequential data compared with earlier neural network architectures.

This keynote will explain how transformers work and why they have become the foundation of today’s frontier AI models. The discussion will begin by describing how text is converted into numerical representations that computational models can process, introducing the concepts of tokenization, embeddings, and positional encoding. It will then explore the attention mechanism, demonstrating how transformers identify the most relevant words and relationships within a sequence rather than processing text strictly one word at a time.

The talk will further explain the key architectural components of transformers, including self-attention, multi-head attention, feed-forward networks, and stacked transformer layers, highlighting how these elements work together to capture meaning, context, and long-range dependencies in language. The processes of training and inference will also be discussed to illustrate how transformers learn patterns from massive text corpora and generate responses by predicting one token at a time.

By focusing on the underlying principles of transformer architectures rather than implementation details, this keynote will provide participants with a solid conceptual foundation for understanding how frontier AI models process language and support an increasingly diverse set of real-world applications.