Updated for 2021 semester 1

A number of projects are possible in all areas, I list the ones I could think of, please also make up your own ideas, and talk me into them. Most of these can be extended to a PhD or reduced for a 3rd year single semester project or a Summer Research Scholarship, or something in between. Also, most of them could range from an AI investigation, to SE construction of a platform, to HCI/IS analysis, and so on.

If you are interested in any of the following (or similar) topics and would like to see if you might want to do a project in this area, please e-mail me tom@cs.anu.edu.au or come and see me in N332.

I suggest you consider my "high priority" projects first. During the COVID lockdown, some of the projects below which mention physiological signals could be done using webcams and computer vision approaches to extract e.g. heart rate. Also, we have lots of data for many of these which has not yet been analysed ...

Projects

Please select the topic and type of project and click the display button.

Year:




Topic:











Type:




 

Bio-inspired Computing

I am interested in bio-inspired computing techniques, and have most expertise in neural networks, fuzzy logic and evolutionary algorithms. I am willing to discuss possible topics using Swarm Intelligence, Ant Trail Optimisation, Immune Networks and so on as well.

MATLAB and hadoop This project is to continue the use of our bio-inspired computing techniques (mostly fuzzy signatures) via hadoop on clusters of commodity hardware. The cluster has been setup over the summer 2014/2015 so its ready for some experiments!

Neural Networks and Deep Learning

I am interested in topics such as extracting rules from neural networks, information retrieval using neural networks, data mining and feature selection, cascade neural network structures, hierarchical neural network structures, and neural network applications. I have published papers in all of these areas with former students so there is plenty of earlier work to build on. Most projects will use the very popular backpropagation neural network training algorithm, and sometime the self-organising map.

Cascade neural network stuctures can be built automatically without making decisions as to the number of neurons required to solve a problem by adding neurons with skip connections. This project could investigate the use of larger chunks such as feature maps as cascade components, which would be useful for recognising images (including faces, or generated art). (Some progress made using shared weight feature-map 'chunks', which can be extended.)
Rule extraction: Neural networks can learn complex tasks but face the problem that human beings do not trust them as they can not understand _why_ a particular decision is made. This project focuses on rule extraction for explanation.
Deep learning for handwriting recognition:
This project involves personalisation of deep learning tools for automated handwriting recognition, starting with existing tools and benchmark datasets and extending to a practical tool.
Deep learning: Various deep learning architectures such as deep neural networks, convolutional deep neural networks, generative adversarial networks, deep belief networks and recurrent neural networks have been applied to fields like computer vision, automatic speech recognition, natural language processing, audio recognition and bioinformatics where they have been shown to produce state-of-the-art results on various tasks. This project is to extend previous work using neural networks for deep learning using eye gaze and physiological data.
Cognitive modeling: The recommendation architecture model is based on a system theoretical approach to understanding higher cognition in terms of physiology. The model has some similarities to conventional neural networks but remains true to biological information flow constraints and does not suffer from the forgetting problem. This project could extend previous work in applications of this model in information filtering and face recognition, or extension of the model to demonstrate some of the properties observed in human consciousness.

Fuzzy Logic

I am interested in topics such as automated construction of fuzzy rule bases from data, hierarchical fuzzy systems, fuzzy interpolation, information retrieval using fuzzy logic, universal approximators, and fuzzy logic applications. I have published papers in all of these areas with former students so there is plenty of earlier work to build on.

* Interval Likert scales and fuzzy intervals: Likert scales are commonly used in HCI to capture qualitative information from users, but can not capture the uncertainty in the mind of the subject. By selecting intervals, this can be captured, and analysed using non-singleton fuzzy systems.
Fuzzy document filtering: Web search engines return many documents which are not relevant to queries. Fuzzy techniques to enhance the results from search engines will be very useful. This project will investigate a number of fuzzy techniques for this purpose.
Combining uncertainty: Fuzzy logic is a technique which deals with uncertainty well. Sometimes data contains multiple kinds or sources of uncertainty. For example, a pedestrian could report that he saw someone he was quite sure was stealing something. He was not certain, and his own reliability is another (different) kind of uncertainty. This project is to develop some methods to combine different kinds of uncertainty in intelligence led investigation. (Data from a company specialising in this area will be available.)
Pattern trees: A pattern tree is a tree which represents the pattern for an output class. The output class is located at the top as the root of this tree. The fuzzy terms of input variables are on different levels of the tree. They use fuzzy aggregations to aggregate from the bottom level to the top (root). For a classification application which involves several output classes, the worked model should have as many pattern trees as the number of the output classes, with each pattern tree representing one class. This project will extend our initial work as published in IEEE Transactions on Fuzzy Logic.
Fuzzy Signature Neural Networks: A neural network model using fuzzy signatures as hidden neurons has shown good results. This project would extend that by finding a quicker automatic way to generate the signatures, to add filtering, and so on.

Evolutionary Algorithms

There are several optimisation methods inspired by natural processes, It has been shown that evolutionary algorithms are efficient tools for solving non-linear, multi-objective and constrained optimizations. The principle is a search for a population of solutions, where tuning is done using mechanisms similar to biological recombination.

Making pictures: If artificial 'organisms' encode the components of an abstract computer generated picture, an individual could identify nice and not nice images repeatedly to generate some 'art' which is tuned for their esthetic sense. Software built here is available to be be extended and improved. This project is ideal if you would like to create Art but do not believe you are artistic (but can recognise something you like).
Scheduling: A previous student has worked on a GA for University exam scheduling. An extension of this work or comparison with bacterial algorithms would be an interesting project. The ANU exam schedule data for last year are available, and the ANU timetabling section is happy to assist.

Human responses in complex interaction projects

I am interested in combining advanced bio-inspired computing techniques and eye gaze and physiological observation of users to model their conscious and non-conscious behaviour during interaction tasks. This area is a meld of my eye gaze / physiological signals / generating art human-dp projects and my neural nets and deep learning / evolutionary algorithms projects. With major funding starting in 2016 and again in 2019, this will be a priority for at least 3 years.

Physiological responses to bad taste: EEG error potentials are known to occur when a person observes or commits an error. This project will investigate whether bad taste or surprise constitutes an error, and whether other physiological or behavioural signals can replace the error potential.
Detecting empathy: We have access to a large dataset of videos classified by degree of empathy shown in the interaction, and have collected a large video dataset of people watching and classifying the degrees of empathy shown. The project would involve computer vision tecchniques on the face videos and on the empathy videos and training a neural/deep AI for prediction and recognition.
Bio-inspired computing and implicit preferences:
Can we infer what a user likes by observing their eye gaze and physiological signals? A number of projects are possible here, ranging from bio-inspired computing approaches to analyse data, to design and running an experiment to collect data.
* Passive and active fatigue monitoring
Our previous results show we can detect stress in from physiological/biometric signals from people, can we also detect fatigue? Does dynamic presentation of information improve or hinder recognition?

NLP projects

I am interested in a few specific NLP topics partly related to teaching.

Question-answer modelling: The use of the Piazza forum is becoming commonplace. In large courses, many of the same questions are asked each year, so an automated question answering tool would be very valuable, and would require innovative research.

Mobile and Wearable Device projects

I am interested in topics such as interactive interfaces for mobile phones or tablets, in the use of novel mobile devices and wearable or unobtrusive devices, and in eReaders.

iOS tools: I am interested in the control of virtual and physical objects via smart phone platforms. Previous work has used iOS devices for generating Art, and for 3D control of virtual objects. The next steps would be either
* Negative bio-feedback: We have a negative biofeedback effectuator, which could be used to subtly and overtly train / control behaviour. This project would involve experiments to collect data and to make statistical or neural network predictions on the data collected.
Mind altering wearables:
We have some biofeedback brain sensor devices such as Muse, Thync, foc.us. A number of projects are possible to collect data in experiments.
eReaders: The use of eInk technology is likely to replace the use of paper books in not too many years. There are few if any scientific comparisons of eReaders. Our current work has produced surprising results, its not always the big companies with the best interfaces. Our next steps would be either
Myo interface: We have some new (pre-release) Myo gesture control armbands. This project will be to extend some programs which use its capabilities, most likely in recognitions of fatigue.

Face Recognition projects

I am interested in topics such as image processing for face recognition, HCI research tool to collect facial images, biologically plausible architectures for face recognition, building and modifying computer models of real faces. No previous work on face recognition is necessary.

Caricature: Comparing a specific face to an averaged face allows us to determine how this face differs from the average. If this difference was increased (an inverse of morphing), we can generate a caricature of the face. (See also 3D face model project, a tool is almost ready.)
Eye gaze: Use of an eye gaze detector and (optionally) an EEG 'hat' allows learning when a user is looking at a face they recognise versus one they do not recognise. If we can tell from eye gaze we have a simple specialised lie detector for the question "do you know this person"! Overall, this use of eye gaze has a number of applications, primarily in the security domain.
View morphing: View morphing between two images of an object taken from two different viewpoints produces the illusion of physically moving a virtual camera. We generate compelling 2D transitions between images. Initially we will use the same face with small rotations and produce intermediate images to produce a realistic QuickTimeVR head.
Deep fake face transfer: Use/extend the recent deep fake tool and similar face transfer tools. Projects could involve extension of such deep learning models, to human experiments to see how well people can detect such fakes.
3-D Face model: An average face 3D model can be constructed from a single face image and animated via rotation, a simple face expression model, or illumination. This can then be used to recognise faces in multiple poses, expressions or lighting. (A tool was built, which can be extended or used for caricature etc.)

Physiological Signals projects

We have a Mind Attention Interface lab, which now also has physiological signal devices such as Galvanic Skin Response, and ElectroCardiogram equipent as well as the Eye Gaze and EEG. Please see the Eye Gaze section also as many of those projects could use the physiological signals to augment or replace Eye Gaze and/or EEG.

Revocable biometrics: Most signals recorded from people could be used as biometric signals for identification or authentication. Like passwords, these should be under our control and revocable at will. Possible solutions include crypotgraphic, blind source separation or machine learning approaches.
Operator observation: Previous projects have built an infrastructure for collecting data in various driving scenarios, and collected data. A project could involve face pose to determine direction of attention or facial expressions.
Driving simulator: we have a driving simulator setup with a force feedback steering wheel with which we have recorded physiological reactions in simulated accidents. Projects in this area could include extending the setup to use eye gaze / EEG or other sensors, or to improve the analysis or ...

Eye Gaze projects

We have a Mind Attention Interface lab, which has Eye Gaze and brain monitoru equipment, and is run in the Interaction lab room. I am particularly interested in interesting uses of the eye gaze equipment, as well as validation using the fNIRS/EEG equipment.

Mac OS Eye gaze tool: Eye gaze trackers are becoming commodity hardware. This project will be to use one of the eye gaze trackers we have to develop a Mac OS program using eye gaze.
Face recognition: see the Eye Gaze project in Face Recognition section
Generating art: see below
EEG/fNIRS: Record EEG/fNIRS and eye gaze while users look at particular kinds of images/scenes. This project could involve:
  1. reliably extract EEG/fNIRS signals known to occur for certain types of images using AI techniques
  2. unsupervised AI techniques for finding new EEG/fNIRS signals
  3. unsupervised AI techniques for characterising eye gaze paths
  4. correlating known/unknown EEG/fNIRS signals with eye gaze path characteristics
  5. software engineering construction of a platform for such work
Attention: what do people see in images? What is it about some images that attract our attention and presumably interest where other seemingly similar images are less interesting?

Smart disk cataloguer and synchroniser

I have many files on multiple computers organised in different ways, and would like to be able to keep track of them, etc. I suspect this is common, especially if we include files on google docs or facebook etc. None of the software I have seen solves my problem. A good solution would be very useful for me and many others.

Web Teaching projects

I am interested in topics related to the way I teach my Web Development and Design course as well as innovative ways of teaching via the web.

3D web: With the 3D MaxWhere tool, for example, a 3D layout of teaching materials would allow a much better understanding of content in context for students. This project would involve development of layouts and conversion of existing materials, to gain insights for the development of a tool to do this automatically.
MOOC tools: I am interested in tools which would assist me in automating tasks which would be impossible for a single or few course staff to do, in settings like a MOOC delivery of the course, possible next steps would be to continue one of:

Generating Art

Some previous work has been done generating Mondrian-like images (see under Evolutionary algorithms). Various projects are possible in this area. These range from the design or implementation of a simulator, to further development of the Mondrian work, to computerised analysis of artistic esthetic, and to generation of other kinds of artistic images.

3D simulator: to show images generated in a 3D visualisation which is either technically interesting or of artistic merit.
Mondrian: a number of further developments are possible here (some initial work done, involved creation of a new drawing tool, which also incorporates some support for a 3D mouse, and initial support for wiimote)
  1. further develop Mondrian image generation process
  2. extend interactive interface
  3. extend to new mobile devices
Rotoscope: Converting live-action film or images to an animated form is called rotoscoping. This project would use existing line detection and optical flow software to automate the conversion of a film to an animated form.
Generate more art: using some other kind(s) of artistic (abstract) images
Faces: modify faces in an identi-kit fashion and present them to users for selection of likeness
Eye gaze: use eye gaze monitor to substitute for manual selection of likes and dislikes (see Eye Gaze section)
Analyse esthetics: can a neural network learn what I find esthetic? (Initial work indicates that this is possible to some extent - this would be interesting to investigate further.)
Reaction: what 3D effects are more powerful? This lends itself best (I think) to an IS/HCI project.