Dr. Ron Sun

Ron Sun, Ph.D.  

Professor and Department HeadCognitive Science Department Rensselaer Polytechnic Institute110 Eighth Street, Carnegie BuildingTroy, New York 12180, USA
Email: Dr.Ron.Sun [at] gmail [dot] com
Web: the Google Scholar pageWeb: the PhilPapers pageWeb: the ResearchGate page
Web: the RPI faculty pageWeb: the RPI homepage   Web: the CogArch Lab page

RESEARCH INTERESTS


My research interest lies in the study and modeling of cognitive agents, especially in their abilities to learn, reason, and act in the real world. More specifically, my research can be roughly categorized into the following main strands: (1) cognitive architectures, (2) connectionist and hybrid (neural-symbolic) models, as well as (3) multi-agent interaction, cognitive social simulation, and cognitive social sciences.


Cognitive Architectures

The goal of the work on cognitive architecture is two fold: to better understand human cognition (broadly defined) in various domains and to develop more unified models (cognitive architectures) for better capturing human cognition generally. Such work thus includes both psychological experiments and computational modeling and simulation. 

A hybrid cognitive architecture Clarion has been continuously developed over the past three decades, which combines both implicit and explicit processes and both procedural and declarative processes in one unified framework. See Clarion

In particular, this cognitive architecture addresses the motivation-cognition-environment interaction, instead of focusing only on cognition in the narrow sense. For instance, it addresses the motivation-cognition interaction, in order to account for the relationship between motivation and performance. It also incorporates emotion, personality, morality, and so on, on the basis of motivation. The model is being used to capture, explain, and simulate a wide variety of relevant human data and phenomena.

This cognitive architecture also addresses the interaction of implicit and explicit cognition. Implicit processes have been shown to have tremendous impact on human cognition and yet they have not been taken seriously into consideration in most cognitive models. Clarion addresses not only implicit processes along side of explicit processes, but, more importantly, also their interaction in learning, reasoning, and various other activities. The resulting model is parsimonious in structure and possesses a variety of learning, reasoning, and other capabilities.

Learning in this cognitive architecture has been particularly concerned with skill acquisition in various domains, ranging from highly intellectual to sensory-motor tasks. It is accomplished, in the main, by reinforcement learning supplemented with rule induction.  The model performs autonomous learning. It develops different types of representations, symbolic and subsymbolic, along side each other. The model has been used to capture, explain, and simulate a wide variety of human skill learning phenomena. See learning.

One of the technical focuses in this regard was the extraction of explicit plans (open-loop policies) based on the results of reinforcement learning, to enable explicit reasoning of plans, without a priori domain knowledge to begin with. A variety of algorithms have been explored for such plan extraction.

Another technical focus was the development of modular reinforcement learning models, in which multiple modules (or agents) compete and cooperate with each other to accomplish tasks, without a priori division of the tasks (i.e., without using any a priori domain-specific knowledge). See multi-agent learning.

Connectionist Reasoning, Knowledge Representation, and Hybrid (Neural-Symbolic) Models

Early on, the research work was mainly concerned with everyday commonsense reasoning by agents. This type of reasoning was characterized by a mixture of rule-based and similarity-based processes, exhibiting both rigor and flexibility. To capture such reasoning, a hybrid connectionist architecture (named CONSYDERR) was developed with both localist and distributed components, that unified rule-based and similarity-based processes and accounted for a variety of CSR patterns. See reasoning.

Within the framework, the following issues were also investigated: (1) The connectionist implementations of rules, logics, and schemas, and the variable binding problem in such implementations. They formed the basis for complex reasoning in connectionist models. (2) Inheritance reasoning, which is an integral part of many CSR patterns. An intensional approach was developed within CONSYDERR that works in constant time. This work suggests that other similar reasoning patterns may also be handled intensionally. (3) Causality, which is an important commonsense construct. A connectionist account was developed based on CONSYDERR, which extended the existing logic-based account and dealt better with the inexact, cumulative, and subjective nature of commonsense causal reasoning. (4) Some attempts have also been made to extend the framework to deal with metaphor and analogy. 

The interest in hybrid models also lies in developing more powerful, more integrated models that are capable of autonomous learning, acquiring both symbolic and subsymbolic knowledge and utilizing their synergy to achieve better performance. Psychological models of human learning were explored, especially those concerned with integrated learning of multiple forms of knowledge, as well as machine learning and neural network techniques and theories, especially those concerned with reinforcement learning. A synthesis of these two strands of work led to advances in developing hybrid models providing new insights and impetus. See learning.

Cognitive Social Simulation and Cognitive Social Sciences

Another general area of interest is: multi-agent interaction, cognitive social simulation, and cognitive social sciences. Extending the work on cognitive architectures beyond individual agents, social interaction needs to be taken into consideration. Social simulation on the basis of cognitive architectures (i.e., cognitive social simulation) enabled the exploration and understanding of many social phenomena in relation to individual cognition. See Cognitive Social Simulation.  See also Cognitive Social Sciences.

Biographical Information

See Wikipedia. See also the old bio here.

ANNOUNCEMENTS

1. To see a description of any of the books, click on a title:

Grounding Social Sciences in Cognitive Sciences
The Cambridge Handbook of Computational Psychology
Cognition and Multi-Agent Interaction

2. To see a description of the past conferences, workshops, and/or journal special issues he (co)organized, click on a title:


3. To see a brief description of our Ph.D program in Cognitive Science, AI and Neural Nets, click here. If you need application forms for the Ph.D program, click here.

Note that I prefer only to supervise graduate students and post-docs who know a lot about my research (i.e., read some of my papers; see below) and wish to do related work. However, I am willing to consider proposals that are slightly afield from exceptionally outstanding students. 

ONLINE RESOURCES

Neural Networks

Connection Science

Cognitive Systems Research  

Cognitive Computation 

2.  HYBRID (NEURAL-SYMBOLIC) SYSTEMS RESOURCES page, which includes:

An introduction to hybrid systems by Ron Sun ("Artificial Intelligence: Symbolic and Connectionist Approaches", an entry in: the International Encyclopedia of Social and Behavioral Sciences. 2001.),

An article by Ron Sun: Hybrid systems and connectionist implementationalism (in: Encyclopedia of Cognitive Science, pp.697-703. Nature Publishing Group (MacMillan). 2002).

Hybrid List moderated by Ron Sun

Bibliography on Connectionist Symbolic Integration (edited by Ron Sun, appeared in the book Computational Architectures Integrating Symbolic and Connectionist Processing, published by Kluwer). 

3. HYBRID REINFORCEMENT LEARNING RESOURCES

4. COGNITIVE ARCHITECTURES RESOURCES

5. COGNITIVE SOCIAL SIMULATION RESOURCES

6. COGNITIVE MODELING RESOURCES at Cognitive Science Society

7. OTHER RESOURCES

SELECTED PUBLICATIONS

Publications by Subject Areas:


Major Journal Papers: 


Books: 


General Overviews: 


Some Other Useful Papers: 


To download the papers, try also the ftp site.