Exploration and Exploitation in Swarm Intelligence
Professor Booncharoen Sirinaovakul, Ph.D.
Department of Computer Engineering,
King Mongkut’s University of Technology Thonburi,
Abstract: Swarm Intelligence is the collective behavior of each member, so called agents or particles, in the swarm. These particles interact with one another and with their environment during their work. Each individual particle follows a set of simple rules and communicates with others on what is the result of it behavior during their interaction. In the meantime, the whole particles in the swarm behave to fine their objective, such as food source. Without any centralized control over how individual particle should behave, the iterations of local interaction among particles and the behavior of the swarm finally lead to the convergence of global behavior of the swarm. Example of Swarm Intelligence in insects such as ant colonies, bee colonies, bird flocking and fish schooling. The interaction locally among particles is so called “exploitation”. On the other hand, the global behavior of the swarm is called “exploration”. In an optimization problem, the balance of exploitation and exploration behavior is the key that leads the swarm to reach the optimal solution. This presentation will show how these two mechanisms of the Swarm Intelligence work.
Booncharoen Sirinaovakul has been working with King Mongkut’s University of Technology Thonburi for over 25 years as a full professor in the Computer Engineering Department. His research interest includes Swarm Intelligence and Optimization. He has been published more than 80 articles in the conference proceedings and international journals. He received a Thailand Frontier Author Awards in 2015 in the field of Computer Science.
Computer Systems and Performance Engineering for Upcoming AI Applications
Associate Professor Yukinori Sato, Ph.D.
Department of Computer Science and Engineering,
Toyohashi University of Technology,
Abstract: Currently, the system-level performance of computer is becoming a big obstacle for realizing the advanced human-centered society driven by upcoming Artificial Intelligence (AI) technologies. For instance, many deep learning programs, which become a typical component for AI-based applications, has struggled with lack of performance even if it is implemented on the state-of-the-art CPUs or GPUs. Therefore, techniques that realize high-performance and high-efficiency computer systems are expected to be an enabler of emerging new AI applications such as self-driving cars and autonomous intelligent robots. In the context of system performance matter, we highlight specialization for inherent memory access locality which will be a clue to its solution. We attempt to improve inefficient part of memory references by fully customizing memory access patterns of application programs towards the target hardware. Here, we propose memory-centric performance engineering that orchestrates various expertise and wisdom of cross-cutting fields of computer systems such as architecture, high-performance computing, programing, software and hardware design. To procreate memory-centric performance engineering, we investigate the followings: (1) memory-centric customization and co-design methodology, (2) scientific modeling of system performance and quality of the performance, (3) automation of customization and co-design driven by mathematical optimization and machine learning techniques. In this presentation, we present our experiences and on-going approaches for memory-centric performance engineering. Finally, we will show the importance of building open ecosystem for highly-customized computer systems for AI applications and resultant AI-inspired principles for computer and mathematical sciences.
Yukinori Sato is an Associate Professor at Toyohashi University of Technology in Japan. His research interests lie in the broad area of high performance computing system, computer system architecture, reconfigurable computing, software and tools for them. In particular, he has been working on developing productive software for parallelism discovery and performance tuning, and a tool for run-time application analysis. He received the Ph.D. degree on Information Sciences from Tohoku University. From 2006 to 2007, he engaged in software engineering for multicore systems and system design of embedded devices in a start-up company. He was an Assistant Professor at Japan Advanced Institute of Technology (JAIST) from 2007 to 2015, and Specially Appointed Associate Professor (Lecturer) at Tokyo Institute of Technology from 2015 to 2018. He is a member of the IEEE Computer Society, the ACM, the IEICE and the IPSJ.