Knowledge heuristics for cognitive computing

Inventing novel means to enable and enhance intelligence in systems by contextual reasoning

Our research

With the advancement in the technology, especially in compute (i.e., advanced distributed graph analytics, such as SPARK and Blazegraph, and chip technology, such as POWER X), applying knowledge-based heuristics to solve complex context-sensitive problems in real time have become a reality. As contextual reasoning techniques can enable systems to mimic human behaviours, especially the thought processes of humans, knowledge-based technologies have become a cornerstone in the cognitive computing era.

Motivated by the above observation, we investigate novel and efficient means to extract, represent, reason, and query-answer on domain-specific corpora by focusing on data semantics. The ethos of our research is to (1) create agile domain models that represent complex domains in an intuitive manner utilising relevant heterogeneous data sources; and (2) apply pragmatic models of human reasoning to effectively and efficiently query-answer on the captured knowledge. We specifically focus on the intersection of natural language processing and formal knowledge representation and reasoning paradigms — both with respect to state-of-the-art machine-learning techniques and Big Data principles — and the impact of pragmatic models of communication for a truly cognitive experience.

From data to knowledge

Pragmatics and discourse in knowledge graphs

With the plethora of available digital information, for example 2.5 trillion PDF documents and the projected footprint of 44 zettabytes of digital products by 2020, the insights gleaned from them are going to truly transform industries in the coming decade and beyond. However, as studies have shown in 2015, only about one-third of this data was useful if analysed, whereas in reality, only 5% of this useful data was in fact analysed.

A little semantics goes a long way.

—Prof. James Hendler, RPI

One of the key problems was due to the rapid growth of unstructured data and the compute requirements for an effective and efficient means to analyse such data in this digital information space. Today, with the advancements in compute and the latest unsupervised learning techniques, we can now realistically explore the true potential of such data for business problems, especially for knowledge-driven decision-making scenarios.

In order to address such scenarios, we are researching a novel set of techniques to develop an end-to-end framework to generate insights from unstructured data and to provide a means for knowledge-driven decision making for a multitude of industries. To support the framework, we are investigating means to

  • combine natural-language processing techniques with Big Data principles to conduct efficient text analytics and machine-learning techniques for topic modelling to automatic term correlation
  • interpret and represent the obtained information as knowledge in an efficient manner by focusing on data semantics [5]
  • query-answer knowledge to support decision making using the latest thought processes in cognitive computing, especially using pragmatic models of communication as illustrated by this data and schema-aware query reformulation pyramid shown here [2, 6].

In the general case, our state-of-the-art algorithms to reformulate queries over large knowledge graphs have been shown to require only half the reformulations needed by the current state-of-the-art for similar results.

The components of this framework have been applied to a number of proof-of-concept implementations from engineering (for example, to provide insights about the thermal properties of the internal combustion engine by analysing patents) to life sciences (for example, to obtain information about biofilms from open-access journals available from the European PMC).

An emerging interest to the group going forward is to look at means to infer high-level semantics about structured data and correlate them with the insight gleaned from unstructured data to start modelling the uncertainty in knowledge so that human-like reasoning can be supported by our framework.

AI at the edge

Semantically-aware resource management

With the proliferation of edge devices, technologies associated with edge computing play a crucial role in the success of critical tasks from production-line monitoring to border surveillance. Since the edge devices and services associated with them are different in capability, only a subset of these resources is useful for a specific task. This is due to the dynamism in the tasks’ environment and the heterogeneity of the resources. Thus, an effective mechanism to select resources for tasks is needed so that the selected resources cater for the needs of the tasks. Though a considerable amount of research has been done in different communities to allocate resources efficiently to tasks, we observed that there is little work being done to guarantee the effectiveness of the selection with respect to the context of operation.

In order to address this problem, we have developed a body of work in which knowledge-based heuristics are used to introduce the context of operation to the resource selection process. Through the use of ontologies, we developed sound and complete mechanisms for effective resource discovery for tasks and an agile means to allocate the required resources to tasks. Our architecture enables a multitude of such knowledge bases to be exposed as services to solve a given problem in a distributed setting.



You can learn more about this by visiting the NIS ITA. To support this work, we have developed a novel, lightweight, and agile knowledge representation framework called ITA Controlled English (CE), which supports deployment of semantic reasoning at the edge.

The research is now being further developed under a $80 million, 10-year basic research programme DAIS ITA, whereby service semantics are inferred through a set of deep and symbolic learning approaches for autonomous service composition in edge systems to satisfy user requirements. This has the potential to revolutionize the service economy for edge computing and we are investigating its applications from smart transportation to intelligent service chaining in IoT infrastructures.

Learn more about this research

Achievement award
For controlled English to support real-time business insight

Achievement award
For scientific advancements in context-sensitive sensor assignment and placement



This is a game-changing tech­nol­ogy be­cause con­text-aware re­source manage­ment pro­vides end users — especially edge users — with the capa­bil­ity to auto­matical­ly engage and se­lect only the most rel­evant sen­sor re­sources for a given mis­sion, which is crit­ical for its success.

—Dr. Tien Pham
Senior Campaign Scientist for Information Sciences
US Army Research Laboratory

Use case: Route planning

Interaction governance for IoT

Autonomous policy generation for connected devices

According to IBM’s vision of a Smarter Planet, efficient networks made up of low-cost devices are going to augment and enhance the day-to-day interactions of humans and organisations. With the proliferation of technology and the Internet, connected and interconnected devices — collectively referred to as Internet of Things (IoT) — this is fast becoming a reality. According to Gartner, there will be over 20 billion interconnected devices by 2020, and they will transform the lives of people and organisations by augmenting their experiences in environments.

With over 50 billion connected devices projected to be active in the coming decade, the challenge is to educate them to learn contextual policies so that they can autonomously govern their behaviours and interactions in evolving environments.

—Dr. Geeth de Mel, IBM Research Staff Member


With the explosion of IoT deployments observed in recent years, manually managing the interactions between humans-to-devices, and especially devices-to-devices, is an impractical — if not impossible — task. This is because devices have their own obligations and prohibitions in context, and humans are not equipped to maintain a bird’s-eye view of the interaction space.

Motivated by this observation, we are investigating technology, especially knowledge-based scalable technology, to develop an agile end-to-end policy framework that

  • automatically discovers devices, and their associated services and capabilities with respect to an ontology,
  • supports representation of high-level and expressive user policies to govern the devices and services in the environment,
  • provides efficient procedures to refine and reason about policies to autonomously manage the interactions, and
  • delegates similar capable devices to fulfil the interactions when conflicts occur [3].

Our current work is focused on mechanisms that enable devices to learn and generate context-sensitive policies and make that learning available to other devices so they can perform similar inferences in matching but previously unencountered contexts [1].

End-to-end policy framework

Ask the experts

David Braines

David Braines
IBM Hursley

Daniel Cunnington

Daniel Cunnington
IBM Hursley

Our collaborators at RPI

Prof. Jim Hendler
Amar Viswanathan

Our collaborator at STFC

Andrew Gargett

Key publications

[1] E. Bertino, G., de Mel, A. Russo, S., Calo, D. Verma,
“Community-based Self Generation of Policies and Processes for Assets: Concepts and Research Directions,”
In Proc. IEEE Big Data workshop on Policy-based Autonomic Data Governance, 2017.

[2] A. Viswanathan, J.R. Michaelis, T. Cassidy, G. de Mel, J. Hendler,
In-context query reformulation for failing SPARQL queries,”
In Proc. Ground/Air Multisensor Interoperability, Integration, and Networking for Persistent ISR VIII, International Society for Optics and Photonics, Vol. 10190, p. 101900M, 2017.

[3] E. Goynugur, G. de Mel, M. Sensoy, K. Talamadupula, S. Calo,
A Knowledge Driven Policy Framework for Internet of Things,”
In Proc. 9th International Conference on Agents and Artificial Intelligence, 2017.

[4] E. Göynügür, S. Bernardini, G. de Mel, K. Talamadupula, M. Şensoy,
Policy Conflict Resolution in IoT via Planning,”
In Proc. Canadian Conference on Artificial Intelligence, LNCS Vol. 10233, Springer, pp. 169-175, 2017.

[5] M. Şensoy, L. Kaplan, G. de Mel,
Semantic reasoning with uncertain information from unreliable sources,”
In Proc. International Conference on Principles and Practice of Multi-Agent Systems, LNCS Vol. 9862, Springer, pp. 92-109, 2016.

[6] A. Viswanathan, G. de Mel, J.A. Hendler,
Pragmatics and Discourse in Knowledge Graphs,” 2015.

[7] B. Donohue, D. Kutach, A. Bhattal, D. Braines, G. de Mel, R. Ganger, T. Pham, R. Rudnicki, B. Smith,
Controlled and Uncontrolled English for Ontology Editing,”
In Proc. CEUR Workshop STIDS, Vol. 1523, pp. 74-81, 2015.

[8] A. Preece, D. Pizzocaro, D. Braines, D. Mott, G. de Mel, T. Pham,
Integrating hard and soft information sources for D2D using controlled natural language,”
In Proc. 15th IEEE International Conference on Information Fusion (FUSION), pp. 1330-1337, 2012.

[9] M. Gomez, A. Preece, M. Johnson, G. De Mel, W. Vasconcelos, C. Gibson, A. Bar-Noy, K. Borowiecki, T. La Porta, D. Pizzocaro, H. Rowaihy,
An ontology-centric approach to sensor-mission assignment,”
In: Knowledge Engineering: Practice and Patterns, LNCS Vol. 5268, pp. 347-363, 2008.