Development of globally optimal solutions to simultaneous localisation and mapping for robot navigation. Building robots that can operate on their own is one of the potentially transformational technologies of this century. This project will develop algorithms that are well understood and robust to allow the deployment of robots in environments populated with people and in search and rescue operations where global positioning system is not available.
An automatic markerless three-dimensional (3D) motion analysis system for aquatic environments. Australia's sporting performance on the international stage forms an integral part of the psyche of Australians. This project applies latest 3D imaging and biomechanical techniques to quantify swimmers' movement patterns, thereby ensuring Australia's continued elite sporting success and consolidating its current lead in world class technologies.
Perceptually-motivated speech parameters for concurrent coding and noise-robust distributed recognition of human speech for mobile telephony systems. With speech being a simple and natural form of communication, speech recognition technology is being widely used in mobile phones. Nowadays, consumers can interact with remote systems via spoken words. This project will develop remote speech recognition with better accuracy and noise-robustness while using the existing mobile phone infrastructure.
Mapping ear morphology to individualised three dimensional audio. The project aims to develop a practical method to derive a listener's individualised Head Related Transfer Functions from two dimensional images of the head and ears. These are essential for generating high-fidelity three dimensional audio. The project will perceptually evaluate and test the proposed system when applied to teleconferencing, surveillance, and navigational guidance.
Deep Learning that Scales. Deep learning has dramatically improved the accuracy of a breathtaking variety of tasks in AI such as image understanding and natural language processing. This project addresses fundamental bottlenecks when attempting to develop deep learning applications at scale. First, this project proposes efficient neural architecture search that is orders of magnitude faster than previously reported, abstracting away the most complex part of deep learning. Second, we will desig ....Deep Learning that Scales. Deep learning has dramatically improved the accuracy of a breathtaking variety of tasks in AI such as image understanding and natural language processing. This project addresses fundamental bottlenecks when attempting to develop deep learning applications at scale. First, this project proposes efficient neural architecture search that is orders of magnitude faster than previously reported, abstracting away the most complex part of deep learning. Second, we will design very efficient binary networks, enabling large-scale deployment of deep learning to mobile devices. Thus this project will overcome two primary limitations of deep learning generally, however, and will greatly increase its already impressive domain of practical application.Read moreRead less
Enhanced Through-Wall Imaging using Bayesian Compressive Sensing. The aim of this project is to develop radar imaging techniques which enable us to 'see' objects behind walls and opaque materials. The major intended breakthrough is the ability to image objects behind walls and inside buildings or enclosed structures without accessing the scene. Novel signal and image processing algorithms, based on Bayesian compressive sensing, will be developed to enhance image quality and resolution, improve s ....Enhanced Through-Wall Imaging using Bayesian Compressive Sensing. The aim of this project is to develop radar imaging techniques which enable us to 'see' objects behind walls and opaque materials. The major intended breakthrough is the ability to image objects behind walls and inside buildings or enclosed structures without accessing the scene. Novel signal and image processing algorithms, based on Bayesian compressive sensing, will be developed to enhance image quality and resolution, improve speed of operation, and reduce the cost and time of data acquisition and processing. Many applications are expected to benefit from this research including search and rescue, surveillance, security, and defence. The research outcomes are expected to enhance the capabilities of the Australian armed forces, counter-terrorism, police and law-enforcement agencies.Read moreRead less
Subband centroids and deep neural networks for robust speech recognition. This project aims to improve the robustness and accuracy of automatic speech and speaker recognition systems. Though these systems work reasonably well in noise-free environments, their performance deteriorates drastically even in the presence of a small amount of noise. To overcome this problem, this project proposes a missing-feature approach for robust speech and speaker recognition. This approach is expected to make th ....Subband centroids and deep neural networks for robust speech recognition. This project aims to improve the robustness and accuracy of automatic speech and speaker recognition systems. Though these systems work reasonably well in noise-free environments, their performance deteriorates drastically even in the presence of a small amount of noise. To overcome this problem, this project proposes a missing-feature approach for robust speech and speaker recognition. This approach is expected to make the speech and speaker recognition systems less sensitive to additive background noise and make them more useful in telecommunications and business.Read moreRead less
Development of methods and algorithms to support multidisciplinary optimisation. This project will aim to develop a number of novel and computationally efficient schemes to deal with the key challenges facing multidisciplinary optimisation. These advancements will allow us to solve a number of challenging and intractable problems in science and engineering.
Efficient learning from multiple brain imaging data sets. Brain imaging data analysis methods have proven to be very effective in the study of brain functions and the identification of brain disorders because they minimise the modelling assumptions on the underlying structure of the problem. Analysis of multiple brain imaging data sets, either of the same modality as in multitask or multisubject data sets or from different modalities as in the case of data fusion, is a challenging problem in bi ....Efficient learning from multiple brain imaging data sets. Brain imaging data analysis methods have proven to be very effective in the study of brain functions and the identification of brain disorders because they minimise the modelling assumptions on the underlying structure of the problem. Analysis of multiple brain imaging data sets, either of the same modality as in multitask or multisubject data sets or from different modalities as in the case of data fusion, is a challenging problem in biomedical image analysis. This project will lead to fundamental contributions as well as techniques that address both problems: extraction of relevant features information from multisubject brain imaging data sets of the same modality or from fusion of brain imaging data sets collected from multimodalities.Read moreRead less
Automated analysis of multi-modal medical data using deep belief networks. This project will develop an improved breast cancer computer-aided diagnosis (CAD) system that incorporates mammography, ultrasound and magnetic resonance imaging. This system will be based on recently developed deep learning techniques, which have the capacity to process multi-modal data in a unified and optimal manner. The advantage of this technique is that it is able to automatically learn both the relevant features t ....Automated analysis of multi-modal medical data using deep belief networks. This project will develop an improved breast cancer computer-aided diagnosis (CAD) system that incorporates mammography, ultrasound and magnetic resonance imaging. This system will be based on recently developed deep learning techniques, which have the capacity to process multi-modal data in a unified and optimal manner. The advantage of this technique is that it is able to automatically learn both the relevant features to analyse in each modality and the hidden relationships between them. The use of deep belief networks has produced promising results in several fields, such as speech recognition, and so this project believes that our approach has the potential to improve both the sensitivity and specificity of breast cancer detection.Read moreRead less