Cognitive Systems (CSG)

What We Do...

Our group conducts research in and develops technologies for Cognitive Systems.

What are Cognitive Systems?

Cognitive Systems are an emerging class of intelligent machines that can interact with humans as adaptive and knowledgeable peers.

Cognitive Systems are inspired and informed by the social and cognitive sciences from which a principled set of built in capabilities are derived. These include:

  • Having basic internal states and drives (such as pain and anxiety) that motivate behaviours from within
  • Possessing background commonsense and world knowledge to guide their behaviours
  • And being able to learn quickly from experience so as to be able to adapt to new environments and tasks.

These capabilities mean that Cognitive Systems do not need to be programmed from scratch for new tasks. Such a system should be able to adapt to new domains through learning; or be programmed starting with the knowledge that it already has.

Research Focus

1. Human-Like Learning

We are developing techniques for human-like learning in machines:
E.g.

  • Learning through experience (e.g. from visual observations)
  • Learning in real-time under real world conditions

One technique that we are working on is causal learning. This aims to achieve fast, experienced-based to be able to learn with very few examples under real world conditions. Contrast this against machine learning (such as reinforcement learning and deep learning) which requires large amounts of training examples and takes a long time to learn.

Selected publications:

  • S. -B. Ho (2012) "The atoms of cognition: A theory of ground epistemics", Proceedings of the 34th Annual Meeting of the Cognitive Science Society, pp. 1685-1690 (Sapporo, Japan)
  • S. -B. Ho (2013a) "A grand challenge for computational intelligence", Proceedings of the IEEE Symposium Series on Computational Intelligence on Intelligent Agents, pp. 44-53 (Singapore)
  • S. -B. Ho (2013b) "Operational representation - A unifying representation for activity learning and problem solving," AAAI 2013 Fall Symposium Technical Reports-FS-13-02, pp. 34-40 (Arlington, VA, USA)
  • S. -B. Ho and F. Liausvia (2013a) "Incremental rule chunking for problem solving," Proceedings of the 1st BRICS Countries Conference on Computational Intelligence (Ipojuca, Pernambuco, Brazil)
  • S. -B. Ho and F. Liausvia (2013b) "Knowledge representation, learning, and problem solving for general intelligence," Proceedings of the 6th International Conference on Artificial General Intelligence, pp. 60-69 (Beijing, China)
  • S. -B. Ho; F. Liausvia, (2014) "A rapid learning and problem solving method: Application to the StarCraft game environment," in Proceedings of the IEEE Symposium on Computational Intelligence for Human-like Intelligence (CIHLI), pp.1-8 (Orlando, Fl, USA)
  • S. -B. Ho (2014) "On effective causal learning," Proceedings of the 7th International Conference on Artificial General Intelligence, pp. 43-52 (Quebec City, Canada)

2. Commonsense Knowledge Representation and Reasoning

We are developing methods for capturing, representing and reasoning with commonsense knowledge in machine readable forms. In particular, we would like to capture narrative knowledge which underlies much of our understanding of the world and provides the basis for humans to have expectations about how things should turn out, instead of treating everything that happens as either chaotic or unrelated.

We aim to make commonsense knowledge available as a resource for reasoning by other developers of cognitive systems.

Selected publications:

  • D Rajagopal, D Olsher, E Cambria, K Kwok (2013) “Commonsense-Based Topic Modeling”, Proceedings of the 19th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD) (Chicago, IL, USA)
  • D Rajagopal, E Cambria, D Olsher, K Kwok (2013) “A Graph-Based Approach to Commonsense Concept Extraction and Semantic Similarity Detection” WWW 2013 Companion (Rio de Janeiro, Brazil)
  • Cambria, E., Poria, S., Gelbukh, A., Kwok, K. (2014) “Sentic API: A common-sense based API for concept-level sentiment analysis”, 4th Workshop on Making Sense of Microposts, Vol. 1141 CEUR Workshop Proceedings, 19–24, (Seoul, Korea)
  • E. Cambria, D. Olsher, and D. Rajagopal (2014) “SenticNet 3: A common and common-sense knowledge base for cognition-driven sentiment analysis”, AAAI, pp. 1515-1521 (Quebec City, Canada)
  • E. Cambria, D. Rajagopal, K. Kwok, and J. Sepulveda (2015) “GECKA: Game engine for commonsense knowledge acquisition”, AAAI FLAIRS, pp. 282-287 (Hollywood, CA, USA)


3. Modelling Human Expertise and Performance

We are developing techniques for computational modelling of human expertise and performance to serve as cognitive/psychological human surrogates in simulations. These models can be applied to HCI/HMI evaluation, facilitate the exploration and understanding of emergent effects of various social policies, and for producing realistic avatars for tutoring systems and interactive applications.




In the longer term, we see Cognitive Systems being applied across different domains, such as in

  • Context-aware applications and games
  • Autonomous systems
  • Social robotics
  • Intelligent Tutoring Systems
  • Intelligent Manufacturing
  • Decision Support Systems, and
  • In providing intelligence and learning capabilities for Internet of Things devices.


Dr. Kenneth Kwok
Principal Scientist