Deep Interaction Learning in Unlabelled Big Data and Complex Systems. This project aims to effectively model intricate interactions deeply embedded in unlabelled big data and complex systems, which are often hierarchical, heterogeneous, contextual, dynamic or even contrastive. Learning such interactions is the keystone of robust data science and for realizing the value of big data but it poses significant challenges and knowledge gaps to existing data analytics and learning systems. The expected ....Deep Interaction Learning in Unlabelled Big Data and Complex Systems. This project aims to effectively model intricate interactions deeply embedded in unlabelled big data and complex systems, which are often hierarchical, heterogeneous, contextual, dynamic or even contrastive. Learning such interactions is the keystone of robust data science and for realizing the value of big data but it poses significant challenges and knowledge gaps to existing data analytics and learning systems. The expected outcomes include new-generation theories and methods for the unsupervised learning of complex interactions in real-life big data, which are anticipated to enable the intrinsic processing of big data complexities and substantially enhance Australia’s leadership in frontier data science research and applications. Read moreRead less
Robust and Explainable 3D Computer Vision. Computer vision is increasingly relying on deep learning which is fragile, opaque and fails catastrophically without warning. This project aims to address these problems by developing new theory in graph representation of 3D geometric and image data, hierarchical graph simplification and novel modules designed specifically for deep learning over geometric graphs. Using these modules, it aims to design graph convolutional network architectures for self-s ....Robust and Explainable 3D Computer Vision. Computer vision is increasingly relying on deep learning which is fragile, opaque and fails catastrophically without warning. This project aims to address these problems by developing new theory in graph representation of 3D geometric and image data, hierarchical graph simplification and novel modules designed specifically for deep learning over geometric graphs. Using these modules, it aims to design graph convolutional network architectures for self-supervised learning that are robust to failures and provide explainable decisions for object detection and scene segmentation. The outcomes are expected to advance theory in robust deep learning and benefit 3D mapping, surveying, infrastructure monitoring, transport and robotics industries.Read moreRead less
Declarative Networks: Towards Robust and Explainable Deep Learning. The aim of this project is to develop declarative machine learning techniques that exploit inherent structure and models of the world. Deep learning has become the dominant approach for machine learning with many products and promises built on this technology. But deep learning is expensive, opaque, brittle and relies solely on human labelled data. This project intends to make deep learning more reliable by establishing theory a ....Declarative Networks: Towards Robust and Explainable Deep Learning. The aim of this project is to develop declarative machine learning techniques that exploit inherent structure and models of the world. Deep learning has become the dominant approach for machine learning with many products and promises built on this technology. But deep learning is expensive, opaque, brittle and relies solely on human labelled data. This project intends to make deep learning more reliable by establishing theory and algorithms that allow physical and mathematical models to be embedded within a deep learning framework, providing performance guarantees and interpretability. This would likely benefit machine learning based products that can understand the world and interact with humans naturally through vision and language.Read moreRead less
Automatic video annotation by learning from web data. This project aims to study next-generation video annotation technologies to automatically tag raw videos using a huge set of semantic concepts. The project will study new domain adaptation schemes and frameworks in order to substantially improve video annotation performance. The resulting prototype system can be directly used by ordinary users worldwide to search their personal videos using textual queries. The system is also applicable to vi ....Automatic video annotation by learning from web data. This project aims to study next-generation video annotation technologies to automatically tag raw videos using a huge set of semantic concepts. The project will study new domain adaptation schemes and frameworks in order to substantially improve video annotation performance. The resulting prototype system can be directly used by ordinary users worldwide to search their personal videos using textual queries. The system is also applicable to video surveillance applications, which can enhance Australia’s homeland security.Read moreRead less
Enabling Automatic Graph Learning Pipelines with Limited Human Knowledge. This project aims to develop an automatic graph learning system for complex graph data analysis. Machine learning for graph data commonly requires significant human knowledge from both domain professionals as well as algorithm experts, rendering existing systems ineffective and unexplainable. This project expects to design novel graph learning techniques which automatically infer graph relations, learn graph models, adapts ....Enabling Automatic Graph Learning Pipelines with Limited Human Knowledge. This project aims to develop an automatic graph learning system for complex graph data analysis. Machine learning for graph data commonly requires significant human knowledge from both domain professionals as well as algorithm experts, rendering existing systems ineffective and unexplainable. This project expects to design novel graph learning techniques which automatically infer graph relations, learn graph models, adapts existing knowledge to new domains, and provide explanations to the graph learning system. The research results should provide benefit to governments and businesses in many critical applications, such as bioassay activity prediction, credit assessment, and drug discovery and vaccine development in response to the pandemic.Read moreRead less
Exploiting Context in Multilingual Understanding and Generation. Automatic translation technologies produce incoherent and incorrect outputs in critical areas, such as health, finance, and law. This is due to translating sentences independently, without regard to the global extra-sentential context and rich linguistic structures inherent in the wider document context. This project aims to exploit global linguistic structures, capitalising on recent advances in deep neural networks, in order to g ....Exploiting Context in Multilingual Understanding and Generation. Automatic translation technologies produce incoherent and incorrect outputs in critical areas, such as health, finance, and law. This is due to translating sentences independently, without regard to the global extra-sentential context and rich linguistic structures inherent in the wider document context. This project aims to exploit global linguistic structures, capitalising on recent advances in deep neural networks, in order to generate coherent and faithful text. Expected outcome include next-generation computational technologies for language understanding and generation. This should significantly benefit document-based language technologies and increase their applications in a range of cultural, industrial, and health settings.Read moreRead less
Deep Weak Learning for Morphology Analysis of Micro and Nanoscale Images. This project will develop novel methods for automated discovery and quantification of image phenotypes from micro and nanoscale images. The outcome will be an advance of the state of the art in biomedical image analysis with a particular focus on generalized weakly-supervised deep learning models for morphological feature representation. The methodologies will transform the deep learning pipeline for real biomedical imagin ....Deep Weak Learning for Morphology Analysis of Micro and Nanoscale Images. This project will develop novel methods for automated discovery and quantification of image phenotypes from micro and nanoscale images. The outcome will be an advance of the state of the art in biomedical image analysis with a particular focus on generalized weakly-supervised deep learning models for morphological feature representation. The methodologies will transform the deep learning pipeline for real biomedical imaging scenarios with high heterogeneity and limited training data. The frameworks will facilitate high-throughput processing for a wide range of microscopy image modalities and biological applications, and potentially become the next generation computational platform to support fundamental research in human biology.Read moreRead less
Bio-inspired Computing for Problems with Chance Constraints. Bio-inspired algorithms have successfully been applied to a wide range of optimisation problems. Uncertainties in real-world applications can lead to critical failures of production schedules or safe critical systems. Chance constraints model such uncertainties and allow to limit the possibility of such failures. This future fellowship builds up the area of bio-inspired computing for problems with chance constraints. It develops high ....Bio-inspired Computing for Problems with Chance Constraints. Bio-inspired algorithms have successfully been applied to a wide range of optimisation problems. Uncertainties in real-world applications can lead to critical failures of production schedules or safe critical systems. Chance constraints model such uncertainties and allow to limit the possibility of such failures. This future fellowship builds up the area of bio-inspired computing for problems with chance constraints. It develops high performing bio-inspired algorithms for stochastic problems where the constraints can only be violated with a small probability. The outcomes will lead to more effective and reliable optimisation methods for complex planning processes in areas of national priority such as mining and manufacturing.Read moreRead less
AUSLearn: AUtomated Sample Learning for Object Recognition. This project aims to enable computers to learn how to effectively use training samples for object recognition. Training sample is the only source used by computers to learn recognising objects. This project creates a new research direction that will enable the first full exploration of the power of samples. The aims will be enabled by leveraging the recent advances in reinforcement learning, fast training algorithms, and by developing n ....AUSLearn: AUtomated Sample Learning for Object Recognition. This project aims to enable computers to learn how to effectively use training samples for object recognition. Training sample is the only source used by computers to learn recognising objects. This project creates a new research direction that will enable the first full exploration of the power of samples. The aims will be enabled by leveraging the recent advances in reinforcement learning, fast training algorithms, and by developing novel deep learning algorithms. The new algorithms will benefit a wide range of applications, e.g. to effectively use car crash training samples for accurately identifying potential road crashes in transport and to effectively use rare medical imaging training data for robustly diagnosing diseases in health.Read moreRead less
A Bayesian Approach to Distributed Estimation for Multi-Object Systems. This project aims to develop new signal processing techniques that facilitate autonomous technologies for environmental perception, with the ability to efficiently process large data volumes from multiple sensing modalities. Rapid advances in sensors and networks have led to a digital data deluge, from which extracting useful information presents new technological challenges and opportunities. To address this development, th ....A Bayesian Approach to Distributed Estimation for Multi-Object Systems. This project aims to develop new signal processing techniques that facilitate autonomous technologies for environmental perception, with the ability to efficiently process large data volumes from multiple sensing modalities. Rapid advances in sensors and networks have led to a digital data deluge, from which extracting useful information presents new technological challenges and opportunities. To address this development, this project seeks to develop new distributed solutions for statistical estimation, which are specifically designed for dynamic systems with multiple object states, and are inherently scalable and robust. The potential benefits include new technologies for smart cities, autonomous infrastructure, and digital productivity.Read moreRead less