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
New Paradigms for Robust Fitting: Kernelisation and Polyhedral Search. Outliers inevitably exist in visual data due to imperfect data acquisition or preprocessing. To enable computer vision applications that can perform reliably, robust fitting algorithms are necessary to counter the biasing influence of outliers. However, current robust algorithms are unsatisfactory: they are unreliable (due to using randomisation) or too computationally costly (due to using exhaustive search). This project wil ....New Paradigms for Robust Fitting: Kernelisation and Polyhedral Search. Outliers inevitably exist in visual data due to imperfect data acquisition or preprocessing. To enable computer vision applications that can perform reliably, robust fitting algorithms are necessary to counter the biasing influence of outliers. However, current robust algorithms are unsatisfactory: they are unreliable (due to using randomisation) or too computationally costly (due to using exhaustive search). This project will develop new robust algorithms to mitigate these shortcomings. It will do so by investigating two new paradigms of kernelisation and polyhedral search, which offer unprecedented theoretical insights into the problem. The outcomes will contribute towards computer vision applications that are more practical and reliable.Read moreRead less
Continuously learning to see. The ultimate goal of computer vision is to make a machine able to understand the world through analysis of images or videos. The new machine learning techniques developed in this project will enable previously impossible methods of computer vision and help strengthen Australia's competitiveness in this important area.
Understanding and Modelling Insect Motion Vision. The interdisciplinary project proposed will offer a stimulating environment for research/training into computational neuroscience, an attractive area for aspiring scientists. We have already demonstrated the feasibility of transferring physiology into applications, and expect this project to deliver functional motion vision models and devices. Our proposed work will have an impact beyond the advancement of neuro-physiology as knowledge gained is ....Understanding and Modelling Insect Motion Vision. The interdisciplinary project proposed will offer a stimulating environment for research/training into computational neuroscience, an attractive area for aspiring scientists. We have already demonstrated the feasibility of transferring physiology into applications, and expect this project to deliver functional motion vision models and devices. Our proposed work will have an impact beyond the advancement of neuro-physiology as knowledge gained is applicable in a range of areas, with applications in miniature unmanned vehicles and collision avoidance detectors in defence and civilian roles. Our project could also assist in the development of artificial intelligence and as a basis for designing implantable artificial eyes.Read moreRead less
Target detection in visual clutter. The interdisciplinary nature of the project will offer a stimulating environment for training a postdoctoral worker in the hot topic of computational neuroscience. While computationally expensive solutions to moving target detection in clutter have been implemented using conventional engineering, this project will offer insight into the efficiency of the biological brain (with benefit of millions of years of evolution towards compact, economical and optimal so ....Target detection in visual clutter. The interdisciplinary nature of the project will offer a stimulating environment for training a postdoctoral worker in the hot topic of computational neuroscience. While computationally expensive solutions to moving target detection in clutter have been implemented using conventional engineering, this project will offer insight into the efficiency of the biological brain (with benefit of millions of years of evolution towards compact, economical and optimal solutions). The results will assist development of efficient artificial intelligence. It will also assist our ongoing collaborations with defence partners to develop and apply algorithms in artificial vision systems. Read moreRead less
Adapting Deep Learning for Real-world Medical Image Datasets. The project aims to investigate new deep learning modelling approaches to leverage real-world large-scale image data sets that contain noisy and incomplete labels and imbalanced class prevalence – to enable the use of these data sets for modelling deep learning classifiers. Expected outcomes include an innovative method for modelling deep learning classifiers. The research will involve new inter-disciplinary and international collabor ....Adapting Deep Learning for Real-world Medical Image Datasets. The project aims to investigate new deep learning modelling approaches to leverage real-world large-scale image data sets that contain noisy and incomplete labels and imbalanced class prevalence – to enable the use of these data sets for modelling deep learning classifiers. Expected outcomes include an innovative method for modelling deep learning classifiers. The research will involve new inter-disciplinary and international collaborations with machine learning and medical image analysis research institutions. This should provide significant benefits, such as better understanding of deep learning theory, new deep learning applications that can use previously unexplored data sets, and training for the future Australian workforce.Read moreRead less
Learning Robotic Navigation and Interaction from Object-based Semantic Maps. Our project aims to develop new learning algorithms that enable robots to perform high-complexity tasks that are currently impossible. Compared to existing methods that rely on low-level sensor data, we aim to achieve this by learning from a high-level graph representation of the environment that captures semantics, affordances, and geometry. The outcome would be robots capable of using human instructions to efficiently ....Learning Robotic Navigation and Interaction from Object-based Semantic Maps. Our project aims to develop new learning algorithms that enable robots to perform high-complexity tasks that are currently impossible. Compared to existing methods that rely on low-level sensor data, we aim to achieve this by learning from a high-level graph representation of the environment that captures semantics, affordances, and geometry. The outcome would be robots capable of using human instructions to efficiently learn complex interaction and navigation behaviours that transfer to unseen environments. Our research should benefit new applications in domains of economic and societal importance that are currently too complex, unsafe, and uncertain for robot assistants, such as aged care, advanced manufacturing and domestic robotics.Read moreRead less
Linkage Infrastructure, Equipment And Facilities - Grant ID: LE130100156
Funder
Australian Research Council
Funding Amount
$210,000.00
Summary
Computational infrastructure for machine learning in computer vision. The many trillions of images stored on computers around the world, including more than 100 billion on Facebook alone, represent exactly the information needed to develop artificial vision. All we need do is extract it. This project will develop the computational infrastructure required to allow Australian researchers to achieve this goal.
Active Visual Navigation in an Unexplored Environment. This project will develop a new method for robotic navigation in which goals can be specified at a much higher level of abstraction than has previously been possible. This will be achieved using deep learning to make informed predictions about a scene layout, and navigating as an active observer in which the predictions informs actions. The outcome will be robotic agents capable of effective and efficient navigation and operation in previous ....Active Visual Navigation in an Unexplored Environment. This project will develop a new method for robotic navigation in which goals can be specified at a much higher level of abstraction than has previously been possible. This will be achieved using deep learning to make informed predictions about a scene layout, and navigating as an active observer in which the predictions informs actions. The outcome will be robotic agents capable of effective and efficient navigation and operation in previously unseen environments, and the ability to control such agents with more human-like instructions. Such capabilities are desirable, and in some cases essential, for autonomous robots in a variety of important application areas including automated warehousing and high-level control of autonomous vehicles. Read moreRead less