Perceptual Attention and Learning (PAL)
Team Members: Asst Prof. Alex Tay
The Perceptual Attention and Learning PAL environment consists of a typical four wheeled, differential steering robotic vehicle that operates and several stationary web-cameras (wall sentinels) connected to a common processing platform. The project is initially set up using conventional control methods to allow the autonomous guiding of the vehicle from a fixed start point to any selected point within the environment with instructions dictated by the wall sentinels.
The initial environment is subsequently replaced with cognitive components developed using perceptual, attentional and autonomous learning concepts
Investigations into Fast Learning Artificial Neural Networks
Team Members: Asst Prof. Alex Tay (PI), M/s Wong Lai Ping (Ph.D. Student. A*STAR SIMTech), Mr Yin Xiang (M.Eng Student, MOE Scholar).
The use of neural networks has been extensive and many applications have been realized over the past 2 decades. Most applications have converged on simple network models, providing a small input vector dimension and solving for well defined output spaces. The difficult problem is however is not in the well defined, low dimensional problems but in high dimensional input spaces with ill defined inputs. The brain is able to filter off ill defined input sequences and generalize onto relevant dimensions. Furthermore, it is consistently able to learn and remain plastic
This investigation studies the Fast Learning Artificial Neural Network (FLANN) model and investigates the possibility of overcoming the curse of dimensionality by separating the input dimension space into sub-spaces that hierarchically recombine as a full network. Investigations into the basic FLANN network also allow the network to be configured as sub-nets within the network. Further work was also done to introduce a Genetic Algorithm type of search capability to stabilize the network initialization structure so that the network can tune itself to type of data set available.
Fundamental Cognitive Vision and Autonomous Navigation Research
Team Members: Asst Prof. Alex Tay, Javier Ibanez Guzman (A*STAR)
Computer vision systems are now able to provide smart applications that address security issues, aid manufacturing processes and save enormous amounts of money. While a standard product parts can be monitored using fixed templates, a patient at the hospital waiting room cannot be easily monitored without latching sensors on the patient. While image recognition and retrieval methods can link images of the same feature contexts, it is still difficult to identify potential terrorists within a public area using simple surveillance systems.
While some may view that computer vision is still progressing well, it is a conviction that there is much promise in the untapped field of cognitive vision. The power in cognitive vision is believed to supersede conventional computer vision in its ability to perceive. The power of perception adds a dimension of inference into the scene extraction, providing a higher level of contextual understanding. Occlusions in the visual spaces are often overcome by the human’s perceptive capabilities. This information extracted with a higher level of contextual content can then be used for the navigation of an autonomous vehicle.
This work is carried out in conjunction with a collaboration between NTU, NUS and SIMTech under the SERC – A*STAR Public Funding Scheme grant for the Collaborative Autonomous Systems for Built Environments (CARSyB) project.
Cognitive Information Systems for Situation Awareness Modelling, Context-Aware Decision Support, and Knowledge Discovery
Team Members: Assoc Prof Tan Ah Hwee (PI), Asst Prof. Alex Tay (co-PI)
Collaborators: Assoc Prof Er Meng Joo (EEE, NTU),
Prof Stephen Grossberg (Boston U.),
Prof Gail Carpenter (Boston U.),
Prof Liz Sonenberg (U. Melbourne)
Situation awareness refers to the perception and understanding of one’s environment for the purpose of planning and critical decision making. This project aims to develop a cognitive architecture for modelling situation awareness. Taking an adaptive learning approach, a self-organizing memory structure will be developed for modelling environment in terms of entities, their attributes with respect to space and time, events, trajectories, and missions. In contrast to traditional situation awareness models, the cognitive memory system will further incorporate a knowledge level for learning critical causal relationships between the actions of the entities, enabling one to perform scenario projection and hypothesis testing. The proposed research can be applied to a wide range of applications, including homeland security for tracking terrorist groups/activities as well as command and control in battlefield modelling and tactical warfare planning.
Cognitive Autonomous Systems
Team Members: Assoc Prof. Ah-Hwee Tan (PI), Dan Xiao (PhD student, part-time)
This project aims to develop a new breed of autonomous systems that possess high level cognitive capabilities, including awareness, robust reasoning, and continuous learning, for operating and adapting in a dynamic environment. Taking an interdisciplinary approach, the project aims to develop biological-inspired adaptive architectures and algorithms by integrating know-how in artificial intelligence, neuroscience, and cognitive science.
Intelligent Technologies for Multimedia Information Fusion and Analysis
Team Members: Assoc Prof. Ah-Hwee Tan (PI), Prof Angela Goh (co-PI),
Assoc Prof Lim Ee Peng (co-PI),
Assoc Prof Clement Chia (co-PI),
Asst Prof Chng Eng Siong (co-PI),
Asst Prof Miao Chun Yan (co-PI), Tao Jiang (PhD student, MOE Scholar), Jiang Xing (PhD student, MOE Scholar)
Collaborators: Dr. Lonce Wyse (I2R)
Information in the ubiquitous media age is typically fragmented and appears in various unstructured and unlabelled forms as data, text, image, audio, and video. This project aims to develop a framework for high-level level analysis and organization across mixed media information. For transforming raw information content into knowledge, the project will develop various cross-media and media-specific technologies for modeling and working with text, audio, images, and video data as well as their unification and association at the semantic conceptual level based on XML schema and MPEG7 standards. To deliver people-oriented rich media content and services to content producers and end users, we further study the issues of user modeling, and develop tools for user- and purpose-centric production, retrieval, organization and repurposing of media content from distributed sources
Smart-Places: UWB-Enabled Context-Aware Service Environment
Team Members: Assoc Prof. Ah-Hwee Tan (PI), Asst Prof. Chuan-Heng Foh (co-PI), Ghim-Eng Yap (PhD student, A*Star Scholar)
Collaborators: Dr. Hwee-Hwa Pang (I2R)
With the continual popularization of UWB technologies and mobile devices,
users in the near future will be able to access applications and services from any location at any time.
The benefits of UWB in high bandwidth transmission and precision localization create opportunities for developing
smart places supported by context-aware mobile applications at locations of strategic importance in Singapore. This project proposes to develop an unified framework that integrates low level (physical) context information, such as locations, time, network condition, and device characteristics, as well as high level contextual information, such as the users’ background, schedule, intention, and commitment. Based on the integrated framework, we will develop adaptive algorithms for learning and refining contextual knowledge in an online and interactive manner.
Adaptive Processing of Data Structures For Image Content Classification, Indexing and Retrieval of Flowers
Team Members: Asst Prof. Cho Siu-Yeung, David (PI)
Collaborators: Dr. Chi Zheru (HK PolyU), Dr. Wang Zhiyong (Univ. of Sydney)
Sophisticated image content representation and classification techniques are a key component in many image processing systems such as trademark registration and verification, digital libraries, medical image archiving, content-based image retrieval, and flower species recognition. Although there are many different methodologies developed for various systems, the performance of the existing systems still does not match that of human visual perception. To develop a sophisticated system that can approach human-like (robust) performance in image retrieval, two main problems should be addressed:
1. Due to error-prone image segmentation process and a large number of variations in which a flower object can appear in an image (some problems derive from the inherent issues of 2D imaging of 3D objects), it is extremely difficult to identify all objects in an image. The second difficulty lies in the representation of spatial relationship among the objects (even after the objects have been correctly identified from an image). In the long run, an interpretation of an image at a semantic level is necessary for a robust image processing and recognition system.
2. Assuming that the image contents can be effectively represented using a data structure such as a tree structure, an immediate question to ask is whether we can find a processing model that can efficiently perform image content classification based on both the structural and statistical information.
SCE SUG
Agents for Personal Mobility
Team Members: Asst Prof. Miao Chun Yan (PI), Li Dongtao (M.Eng student)
Collaborators: Prof Yang Qiang, HKUST
Today, a large number of users already utilize a wide variety of mobile devices ranging from simple mobile phones and personal digital assistants (PDA) to high-end multimedia notebooks. Personal mobility is the key challenge to providing anytime, anywhere services in any form. In this research service oriented architecture as well as a framework for using mobile and intelligent agents to support personal mobility are explored
Agent Mediated Service Oriented Grid
Team Members: Asst Prof. Miao Chun Yan (PI), Weng Jianshu (PhD student), Peter Leung (PhD student, Part-time)
Collaborators:
Dr. Cao Jun Wei, Data Grid, MIT
There have been increased research interests in the context of Service Oriented Grid, such as service discovery, service negotiation and service provision etc. Negotiation is a vital component that facilitates the Service Oriented Grid as well as Market-Oriented Grid. Negotiation agents play an indispensable role within the Service/Market Oriented Grid. Although agent negotiation has been explored in some application domains, especially in E-Commerce, the extremely dynamic nature of grid makes agent negotiation a new challenging research issue.
The objective of this research is to explore a particular type of intelligent agents, namely, dynamic cognitive negotiation (DCN) agents for facilitating service negotiations in service oriented grid.
|
