[Legacy Report] SMC and SP Seminar on Symmetrical Hierarchical Stochastic Searching on the Line in Informative and Deceptive Environments
IEEE Northern Jersey Section SMC and SP Chapters
Symmetrical Hierarchical Stochastic Searching on the Line in Informative and Deceptive Environments
Junqi Zhang, Ph.D. and Associate Professor
Department of Computer Science and Technology, Tongji University, Shanghai, China,
Time: 2pm, Nov. 13, 2015
Place: ECE 202
Abstract:A stochastic point location (SPL) problem aims to find a target parameter on a one-dimensional line by operating a controlled random walk and receiving information from a stochastic environment (SE). If the target parameter changes randomly, we call the parameter dynamic; otherwise static. SE mainly consists of two types of environments—informative (p > 0.5 where p represents the probability for an environment providing a correct suggestion) and deceptive (p < 0.5). Up till now, hierarchical stochastic searching on the line (HSSL) is the most efficient algorithm to catch static or dynamic parameter in an informative environment when 0.6181 < p < 1, but it is unable to locate the target parameter in a deceptive environment and unable to recognize an environment’s type (informative or deceptive). This paper presents a novel solution, named symmetrical HSSL (SHSSL), by extending an HSSL binary tree-based search structure to a symmetrical form. By means of this innovative way, the proposed learning mechanism is able to converge to a static or dynamic target parameter in the range of not only 0.618 < p < 1, but also 0 < p < 0.382. Finally, the experimental results show that our scheme is efficient and feasible to solve the SPL problem in both informative and deceptive environments.
Biography:
Junqi Zhang received his Ph.D. in Computing Science from Fudan University, China, in 2007. He became a Post-Doctoral Research Fellow and a Lecturer at the Key Laboratory of Machine Perception, Ministry of Education in Computer Science at Peking university, Beijing, China in 2007, and was the recipient of the Outstanding Post-Doctoral Award from Peking university. He is currently a Research Fellow and an Associate Professor with Department of Computer Science and Technology, Tongji University, Shanghai, China. He is a visiting scholar in NJIT from Nov. 2014 to Nov. 2015. He is a Senior Member of IEEE. His current research interests include machine learning, intelligent and learning automata, particle swarm optimization, high-dimensional index and multimedia data management.
Contact: Prof. Mengchu Zhou at zhou@njit.edu if any question. ECE 202 is located at the intersection between warren St. and Summit St., Newark, NJ 07102.
Add Event to Calendar
