Artificial Intelligence Research Group (AIRG)

AIRG Mission

To be an AI research group of international standing and vision, building a rich and stimulating research culture in research training and collaboration, and achieving research excellence through the development of theoretical foundations and innovative technologies for complex AI systems.

AIRG Research Focuses

  • Machine learning and AI photography: image-to-image translation using generative adversarial networks, image and photograph aesthetic assessment using deep learning.
  • Knowledge representation and reasoning:  Ontology based reasoning and data query; computational properties of existential rules; knowledge-based reasoning with Big Data; Answer Set Programming (ASP).
  • Intelligent agents: automated negotiation; reasoning about strategy; trading agents.
  • Machine learning: image processing; AI-based real time analysis and self-adaption.
  • Information security: network security; secured data access in shared cloud storage; rule-based access control policies.

AIRG Contact

Professor Yan Zhang:

Current AIRG Research Projects

Project title: Image-to-Image Translation with Merging Generative Adversarial Networks

Research areas: Artificial Intelligence, Deep Learning, Generative Adversarial Network, Image Recognition, Neural Networks, AI and Art

Many problems in image processing and computer vision can be viewed as image-to-image translation where an image from a domain is translated into another image in a different domain. In most of existing works in this area, both input domain and output domain are known, the translation becomes a problem of finding a mapping between these two domains. In this project, we consider a much more general situation: given two input domains and two images from these two domains, respectively, we want to merge these two images, and generate a new image which from some unknown domain, while such output images should be meaningful in terms of some abstract concepts and features, such as infinity, loop, embeddedness, etc, from an artistic viewpoint.

In this project, we will develop a novel architecture of generative adversarial networks which can be applied as a generic GAN model to merge images from different domains and then generate new images which belong to some undefined new domain. The output mages from MergeGAN should be of sufficient resolution and details and able to be further expended. To apply MergeGAN for image merging, users do not need to do any hard-coding, just use different training datasets.

Key people: Yan Zhang and Vernon Asuncion


Project title: Intelligent Photography Curation Using Deep Learning

Research areas: Artificial Intelligence, Computational Photography, Deep Learning, Image Recognition, Neural Networks

As one of the most predominant machine learning approaches, Deep Learning has been widely applied in various complex image classification and recognition tasks. One such task is to automatically curate photographs taken by human, based on Human’s aesthetic criteria. However, the main challenge for doing this is how to develop a system which can learn such aesthetic criteria. In recent years, some research works have been done in this area, but they generally have major limitations in one way or another.

The aim of this project is to develop an advanced photograph curation system, by using the deep learning approach. By formalizing critical aesthetics in photography, using precise and declarative logic and mathematic formulas, the system will employ a designated deep neural network to learn such aesthetic criteria, such as colours, lighting, compositions and creativity, from many large photography datasets. It is expected that resulting system will perform serious photo curation tasks at the level of professional art curators.

We need to highlight that Professor Yan Zhang is not only an international leading AI researcher, but also an international award-winning landscape photographer.  This is an ideal research project bringing Art and AI together.

Key people: Yan Zhang and Vernon Asuncion


Project title: Spatio-Temporal Data Integration Platform for Information Sharing and Data Mining

This project aims to develop a data integration platform to create a data source for evaluating environmental (climate, pollution and demographic) impacts on many relevant application domains in spatiotemporal scale. Such applications currently under investigation include health and disease, regenerative medicine and biotechnology, criminology and crime, and agriculture and biosecurity. Aims to, study the impacts of environmental changes to obtain vulnerability profiles, then to translate to policies for managing and mitigating burden under consideration such as Asthma, Diabetes, Cataract, and Crime in real time. Open source software is been used.

Spatiotemporal data is large in volume and dependency patterns are more complex and changes with time. Another issue is spatial heterogeneity and temporal non-stationarity of data. This platform provides a data source to develop and test new data mining techniques and algorithms to analyse such spatiotemporal data and then to use for real time predictive modelling.

Key people: Liwan Liyanage, Roshan Yapa, Michael O’Connor and Karl Roberts

Contact: Liwan Liyanage –

Project title: Enhancing Answer Set Programming with Deep Learning

Research areas: Artificial Intelligence, Knowledge Representation and Reasoning, Machine Learning, Neural Networks

Project description:

Learning and reasoning are two most important features of Artificial Intelligence (AI). In recent years, research on both machine learning and knowledge representation and reasoning has made significant progress in both theories and applications. In particular, deep learning has become one of the predominant machine learning approaches and has applied in many real-world domains, while knowledge representation and reasoning has developed expressive and efficient systems for various problem-solving tasks.

This project aims to develop a formal foundation and a practical system to integrate learning and reasoning for complex problem solving. In particular, by enhancing the Answer Set Programming (ASP) – the main programming language for knowledge representation and reasoning, with probabilistic extensions, we will be able to perform dataset training in deep neural network learning with ASP-based reasoning capability.

The project contains three major research components: extending ASP with probabilities over atoms, which are treated as training outputs from neural networks; developing a practical solver for this extended ASP; and undertaking extensive experiments for various learning and reasoning tasks.

Key people: Vernon Asuncion and Yan Zhang


Project title: Artificial intelligence enhancing Blockchain technology

Blockchain technology provides a new concept for data management in current digital era, where data sharing, auditing, access and query answering are the key components. This project aims to use the advanced AI methodologies and technologies to enhance Blockchain based data management.

In particular, this project has three major goals. Our first goal is to develop a Blockchain based data integrity verification (auditing) in P2P cloud storage, making verification more open, transparent, and auditable. Under this framework, we plan to apply Merkle tree for data integrity verification, and analyze the performance of the system under the different Merkle tree structures. New sampling strategy and verification algorithms will be proposed and developed; smart contract will be then designed and implemented in the Blockchain to fulfill this verification process. We will also develop a number of real world applications based on this Blockchain based data integrity verification, including data matching for properties, insurance compliance checking, and auditing.

Blockchain can be viewed as a distributed database where transactions are stored in a ledger which all involved parties hold a copy of. Blocks are time stamped batches of valid transactions. In this Blockchain environment, efficient query answering is a challenging issue. Our second goal of this project is to develop an effective query answering mechanism for Blockchain systems, in which we will have a unified query language suitable for different levels of queries, it is efficient for handling Big Data, and is able to express complex knowledge/ontology based queries.

The third goal of this project is to develop a Blockchain based data access control system. The traditional data access control models have been suffering from the following problems: one point of failure; a user’s access to a resource is not publically known, which leaves space for malicious’ attacks; and most traditional access control models are centralized. Blockchain based access control model will completely overcome all these drawbacks, and provide more secure and robust data access. Advanced answer set programming (ASP) approach will be used in this development.

Key people: Yan Zhang, Simeon Simoff, Vernon Asuncion, Dongmo Zhang, and Yi Zhou

Partners: IBM (Australia), Intech Solutions

Contact: Yan Zhang -

Project title: Building robust network topologies against cyber attacks

Network topology (i.e., how nodes are connected by links in a network) is vital for the operation of networks. Cyber attacks to bring down the nodes or links in networks have surged in recent years. By leveraging the Complex Networks theory, which is widely used to study complex real-world networks, this project aims to (1) design network topologies that are robust against cyber attacks under budget constraints, (2) develop new metrics for measuring and analysing network robustness such that we can use them as guidelines to enhance network robustness and (3) contribute the software tools in achieving the previous two aims as open source software.

The outcomes of this project can be broadly applied to infrastructure networks such as enterprise backbone networks, water/electricity distribution networks, transportation networks, IoT network for smart cities, etc

Key people: Weisheng Si, Yun Bai, Weixing Zheng

Partner: University of Sydney

Funding organization: NSW State Government

Contact: Weisheng Si -

Project title: Multi-robot systems for autonomous driving

This project aims to develop automated negotiation technologies for autonomous driving. We investigate the protocols for autonomous vehicles to communicate and negotiate each other for sharing public roads and avoiding collisions. We treat autonomous vehicles as robots so that we can develop software systems based on the Robot Operating System to implement and test negotiation protocols. 

Key people: Dongmo Zhang, Dave de Jonge, Jianglin Qiao and Michael Abdo

Contact: Dongmo Zhang -

Project title: Efficient and fairer assessment or training systems via AI-based real-time analysis and self-adaption

Adaptive skill training and efficient and fair assessment systems are not foreign to AI-based techniques and strategies. Accurate and fair assessments on lengthy and complex student work, for instance, can be very tedious and draining due to the possible infinite answer variations and the corresponding need to scale their similarities and to repeat a good portion of comments or feedback, especially when some have to intensively assess many scores of such works to meet the deadlines. The assessment consistency becomes even more challenging when a relatively larger group of assessors are required to assess the work on their own and yet to produce the outcome of the same standard. Here we propose to track all students’ marks and feedback comments by all assessors and on all parts of students’ prior and current work real-time, analyse and predict the similarities, and prompt for most similar work assessed so far in terms of their received marks and comments. This will first of all lead to more uniform and therefore fairer assessments across different students and different assessors. Secondly it could provide predicted marks and feedback, pending potential rectification and modification, to greatly improve the assessment efficiency and consistency. The training system, on the other hand, can be built along a similar principle in that automated training exercises can be analysed real-time to alter the training content and complexity so as to best enhance the training effectiveness.

Key people: Zhuhan Jiang, and Jiansheng Huang

Contact: Zhuhan Jiang -

Project title: Loop restricted existential rules and first-order rewritability for query answering

In ontology-based data access (OBDA), the classical database is enhanced with a theory called ontology in the form of logical assertions generating new intensional knowledge. A powerful form of such logical assertions is the tuple generating dependencies (TGDs), also called existential rules, where Horn rules are extended by allowing existential quantifiers to appear in the rule heads. In this project, we introduce a new language called loop restricted (LR) TGDs (existential rules), which are TGDs with certain restrictions on the loops embedded in the underlying rule set. We study the complexity of this new language. We show that the conjunctive query answering (CQA) under the LR TGDs is decidable. In particular, we prove that this language satisfies the so-called bounded derivation-depth property (BDDP), which implies that the CQA is first-order rewritable, and its data complexity is in AC0. We also prove that the combined complexity of the CQA is EXPTIME-complete, while the language membership is PSPACE-complete. Based on LR TGDs, we will extend the LR TGDs language to the generalized loop restricted (GLR) TGDs language. We will prove that this class of TGDs still remains to be first-order rewritable and properly contains all of other first-order rewritable TGDs classes discovered in the literature so far.

Key people: Yan Zhang, Vernon Asunciuon, and Heng Zhang

Contact: Yan Zhang -

Project title: Logical foundations and applications of general intelligence

This project aims to develop the fundamental techniques that enable systems to understand descriptions of new tasks and to generate or learn their own strategies for the tasks effectively without human intervention. The investigation will be conducted in alignment with the researches on general game playing (GGP), which aims to build autonomous agent capable of playing any formally described games. The project also aims to apply the techniques of general problem solving to business automation, automated negotiation and electronic trading.

Key people: Dongmo Zhang, Michael Thielscher, Dave de Jonge, and Siri Padmanabhan Poti

Contact: Dongmo Zhang -

Project title: Automated IQ test

AI benchmarking becomes an increasingly important task. As suggested by many researchers, Intelligence Quotient (IQ) test, which is widely regarded as the standard benchmark for testing human intelligence, is one of the best candidates for testing machine intelligence as well. Nevertheless, most existing IQ test datasets are not comprehensive enough for this purpose. As a result, the conclusions obtained are not representative. To address this issue, this project aims to create a large-scale IQ test dataset and to promote it as a standard AI benchmark. Research on automated IQ test cover many aspects of advanced AI research areas including knowledge representation and reasoning, machine learning, natural language processing, image understanding.

Key people: Yi Zhou, Zhiyong Feng, Guozheng Rao and Haodi Zhang

Contact: Yi Zhou -

Project title: Answer set programming with function symbols

Answer set programs with function symbols may have infinite stable models in general, and as such, the traditional ASP solvers cannot ground such programs. A program is called finitely ground if there is a finite propositional logic program whose stable models are exactly the same as the stable models of this program. Therefore, finite groundability is an important property for logic programs with function symbols because it makes feasible to compute such programs' stable models using traditional ASP solvers. In this project, we explore new decidable classes of finitely ground programs

from polynomial bounded to exponential bounded programs, so that the new discovered classes will strictly contain all other existing decidable classes of finitely ground programs. We also study the relevant complexity properties for these classes of programs. We also investigate various applications of these new bounded ASP programs in complex domains such as planning, and implement practical solvers handling these bounded programs in effective ways.

Key people: Vernon Asuncion, Yan Zhang, and Heng Zhang

Contact: Vernon Asuncion -

Project title: The triangularly guarded class of existential rules

In our recent research on existential rules in ontology based data access (OBDA), we introduced a new class of tuple-generating dependencies (TGDs) called triangularly guarded TGDs, which are TGDs with certain restrictions on the atomic derivation track embedded in the underlying rule set. In this project, we study the decidability of conjunctive query answering under this new class of TGDs. We will further show that this new class strictly contains some other important TGDs classes such as weak-acyclic, aGRD, guarded, sticky and shy. In this sense, we will demonstrate that this triangularly guarded class of TGDs provides a unified representation of all these aforementioned classes. Algorithms and prototype of checking the membership of this new class of TGDs will be also developed.

Key people: Vernon Asuncion and Yan Zhang

Contact: Vernon Asuncion -