[Picture of Maurice!]

Professor Maurice Pagnucco

Deputy Dean (Education)
Dean's Unit, Level 6, Building K17,
Faculty of Engineering,
University of New South Wales,

Telephone: +61 2 9385 5000
Email: morri@cse.unsw.edu.au
WWW: http://www.cse.unsw.edu.au/~morri/
LinkedIn: https://www.linkedin.com/in/mauricepagnucco/?originalSubdomain=au

Maurice Pagnucco Research Projects

I am particularly interested in logic-based approaches to artificial intelligence and their applications, particularly to robotics. Here is a list of suggested topic areas. Also, if you're interested in these areas and would like to propose or discuss an alternative project, please contact me.

The following projects can be offered at PhD, Masters by Research, Postgraduate Coursework Masters research project or Honours/Thesis level. I have grouped them into several main themes:

Ethics and AI

Computational Modelling of Ethical Behaviours

With Dr Yang Song.

Ethical decision making is critical in autonomous machines, such as self-driving cars and social robots. They need to follow social and behavioural norms when performing actions in order to be widely accepted into society. Currently various approaches have been developed for ethical decision making, ranging from logical reasoning to deep learning. In this project, we will develop a formal approach to model ethical principles and an experimental platform to evaluate ethical decision making. Specifically, we will design several real-life scenarios that demonstrate different aspects of ethics, develop logical reasoning models to encode ethical decision making, and conduct experimental studies of the developed model using real-life scenarios. As a further study, methods based on reinforcement learning will also be investigated for comprehensive comparison between logic-based and learning-based approaches. This study will contribute towards AI ethics, which is an emerging research area in the AI community and industry.

Representing and Reasoning with Ethical Principles in AI Programs

Current advances in artificial intelligence (AI) have led to a great interest in ensuring that AI programs adhere to certain ethical principles or behavioural norms. This project will explore an approach to this issue using the experimental cognitive robotic programming language Golog. It will be implemented on one of the robot platforms in the CSE robotics lab: Softbank Nao or Toyota HSR.

Symbolic and Sub-Symbolic AI

Explainable AI with Statistical Models

With Dr Yang Song.

Deep learning models are typically considered as a black-box while explainability is important for critical applications such as autonomous driving and medicine. Explainable AI has thus become an important topic in research and industry. Various methods for explaining deep learning models have been developed, including visualisation-based techniques and feature relevance evaluation. In this project, we will investigate the relationship between deep learning and statistical models. Current research has demonstrated that it is possible to use deep learning to simulate statistical models or vice versa. However, we are interested in further investigation of the underlying formulations and mathematical equivalence between these models. This project will involve extensive literature review and in-depth analysis of various deep learning and statistical models, and conduct experiments on classical benchmarks to derive fundamental understanding of the model formulations.

Integrating Symbolic and Sub-Symbolic Artificial Intelligence

With Dr Yang Song.

Over recent years, machine and deep learning techniques have shown great success in many applications, such as computer vision and natural language processing. However, in many cases purely data-driven approaches would provide suboptimal results especially when not enough high-quality data are available for training the models. Moreover, we hypothesise that prior-knowledge, such as logic rules and knowledge graphs, if integrated with the learning-based models can provide improved performance with a smaller amount of training data.

In this PhD project, we will investigate novel methods that effectively integrate knowledge representation and reasoning with machine/deep learning, which are the symbolic and subsymbolic approaches of artificial intelligence. We will also investigate the novel applications of the designed methods in various domains, such as computer vision, human-robot interaction and autonomous driving.

Deep Learning in Social and Interactive Environments

With Dr Yang Song.

Visual information from images and videos can provide important social signals, such as social relationship, emotion and intention, which affect the downstream tasks such as movement and behaviour prediction. Predicting human movement, for instance, enables robots to conduct in a socially acceptable manner. This new research area, artificial social intelligence, covers a wide range of research topics that typically require semantic analysis, graphical modelling, video analysis and generative models, which are key areas in computer vision and deep learning. In this project, we will focus on a particular research topic in artificial social intelligence and develop novel deep learning based approaches. Example topics include path prediction, social relationship recognition, action recognition, behaviour prediction and human-object interaction. Addressing these problems is a crucial step for many autonomous platforms such as self-driving cars, social robots and surveillance.

Automated Disassembly of End-of-Life Products

A Generic and Robust Ontology and Knowledge-Base System for Robotic Disassembly of E-Waste

Current practises in treatment of hazardous end-of-life products such as electronic waste (or e-waste) include disposal in landfill, shredding for recycling or manual disassembly. They come with a variety of issues such as having detrimental effects on the environment, lacking safety and ethical regulations in developing countries, or failing to efficiently utilise remaining energy and resources embodied within the product. A cognitive robotic disassembly system which addresses these issues whilst being able to handle uncertainties prevalent in end-of-life products requires an ontology or other means of storing knowledge and relationships to reason about. This project will involve developing, testing and evaluating a generic but robust ontology and knowledge-base system of products and their components, fasteners, joining methods, disassembly actions and tools.

Machine Learning Computer Vision for Robotic Disassembly of E-Waste

Current practises in treatment of hazardous end-of-life products such as electronic waste (or e-waste) include disposal in landfill, shredding for recycling or manual disassembly. They come with a variety of issues such as having detrimental effects on the environment, lack safety and ethical regulations in developing countries, or fail to efficiently utilise remaining energy and resources embodied within the product. A cognitive robotic disassembly system which addresses these issues whilst being able to handle uncertainties prevalent in end-of-life products requires a computer vision system capable of correctly identifying various components and fasteners such as screws. This project will involve developing a substantial enough dataset of images for training and applying machine learning computer vision techniques for screw detection. Various machine learning algorithms will need to be explored to select the most appropriate methods.

Knowledge Represenatation and Reasoning

Building a Meta Learning Management System

There have been many Learning Management Systems (LMS) developed for providing students with course materials, guidance through a course, etc. Examples include Moodle (which is used by UNSW), WebCMS developed in CSE, OpenLearing (developed in CSE), Smart Sparrow (developed in CSE), Canvas, etc. One issue that is not completely addressed by these systems is the ability to easily repurpose and reuse digital assets and content. Most provide a way of using course content from a previous offering of a course however this only provides a limited amount of flexibility. This project proposed building a Meta LMS. That is a learning management system that focus on a more fine-grained view of education resources, digital assets, contents, etc. and provides the ability to select from these resources to populate an LMS.

RoboCup Soccer and @Home

With Professor Claude Sammut and Dr Yang Song.

UNSW's rUNSWift RoboCup soccer team and sUNSWeep RoboCup@Home team compete in the annual RoboCup international robotics competition. I can propose several projects working with these teams.

Social Applications of AI

Deep Learning for Indigenous Arts

With Dr Yang Song.

Effective presentation of indigenous arts can promote culture conservation and education. While there is widespread interest in indigenous arts among the general public, it is difficult for the English-speaking community to fully appreciate indigenous arts due to language and cultural barriers. In this project, we will develop several deep learning-based applications that can help break the barriers. In particular, we will develop a system that can translate between English and certain common indigenous languages. As a further study, we will develop a system that can recognise indigenous texts from photos/images and then translate them into English. We will also investigate edge computing for deployment of such systems as mobile applications. The project will involve data collection, development of deep learning models, experimental analysis and application development. Extended studies will include visual arts into the applications, such as image-text retrieval and neural style transfer.

Open-source Demonstration of Common AI Techniques

With Dr Yang Song.

In this project, we will develop a series of open-source packages and tutorial materials to demonstrate the use of artificial intelligence (AI) methods in application development. While our students have studied multiple AI courses, it is often difficult to apply the learned knowledge to real project development. Although lots of online tutorials are available, selecting the best material to learn from has been challenging. It would be ideal to develop a collection of open-source packages to provide students with some baselines or samples, and also develop blog/video tutorials of some well-known AI methods. In this project, we will develop these materials to demonstrate a variety of AI techniques including deep learning for computer vision and natural language processing, knowledge representation and symbolic reasoning, and statistical modelling and machine learning. Such resources will help students quickly get up to speed in their project development. A reference for this project is: Machine Learning is Fun.

Maurice Pagnucco (morri@cse.unsw.edu.au)
PGP/GPG Public Key