Towards knowledge discovery from imperfect and evolving data. Information extraction from data is critical, both to analyse and protect consumer data. However, many learning techniques are developed using perfect, static datasets, quite different to messy, ever-changing real-world data. This project aims to develop data analytics techniques that can extract accurate information in complex structures from imperfect/incomplete data that changes over time. Expected outcomes are a prototype tool, te ....Towards knowledge discovery from imperfect and evolving data. Information extraction from data is critical, both to analyse and protect consumer data. However, many learning techniques are developed using perfect, static datasets, quite different to messy, ever-changing real-world data. This project aims to develop data analytics techniques that can extract accurate information in complex structures from imperfect/incomplete data that changes over time. Expected outcomes are a prototype tool, tested on real datasets, that combines new techniques in data modelling, algorithm development, and system design. Likely benefits are enhanced Australia's competence in data science through student training and new, robust data tools relevant to critical sectors such as cybersecurity, healthcare, and defence.Read moreRead less
Short Sequence Representation Learning with Limited Supervision . Predicting events based on short text and video data is widely found in real-world applications such as online crime detection, cyber-attack identification, and public security protection. However, to develop such an effective prediction model is very difficult due to the problems such as limited supervision, heterogeneous multiple sources, and missing and low-quality data. This project is to tackle these challenges. Expected outc ....Short Sequence Representation Learning with Limited Supervision . Predicting events based on short text and video data is widely found in real-world applications such as online crime detection, cyber-attack identification, and public security protection. However, to develop such an effective prediction model is very difficult due to the problems such as limited supervision, heterogeneous multiple sources, and missing and low-quality data. This project is to tackle these challenges. Expected outcome of this project will lay a theoretical foundation for effective short sequence representation learning and build next-generation intelligent systems. This should benefit our society and economy through the applications of multimodality-integrated video technologies for cybersecurity and public safety. Read moreRead less
Small Scalable Natural Language Models using Explicit Memory. Deep neural networks have had spectacular success in natural language processing, seeing wide-spread deployment as part of automatic assistant devices in homes and cars, and across many valuable industries including finance, medicine and law. Fueling this success is the use of ever larger models, with exponentially increasing training resources, accompanying hardware and energy demands. This project aims to develop more compact models ....Small Scalable Natural Language Models using Explicit Memory. Deep neural networks have had spectacular success in natural language processing, seeing wide-spread deployment as part of automatic assistant devices in homes and cars, and across many valuable industries including finance, medicine and law. Fueling this success is the use of ever larger models, with exponentially increasing training resources, accompanying hardware and energy demands. This project aims to develop more compact models, based on the incorporation of an explicit searchable memory, which will dramatically reduce model size, hardware requirements and energy usage. This will make modern natural language processing more accessible, while also providing greater flexibility, allowing for more adaptable and portable technologies.Read moreRead less
Learning to Reason in Reinforcement Learning. Deep Reinforcement Learning (RL) uses deep neural networks to represent and learn optimal decision-making policies for intelligent agents in complex environments. However, most RL approaches require millions of episodes to converge to good policies, making it difficult for RL to be applied in real-world scenarios taking significant resources. This project aims to equip RL with capabilities such as counterfactual reasoning and outcome anticipation to ....Learning to Reason in Reinforcement Learning. Deep Reinforcement Learning (RL) uses deep neural networks to represent and learn optimal decision-making policies for intelligent agents in complex environments. However, most RL approaches require millions of episodes to converge to good policies, making it difficult for RL to be applied in real-world scenarios taking significant resources. This project aims to equip RL with capabilities such as counterfactual reasoning and outcome anticipation to significantly reduce the number of interactions required, improve generalisation, and provide the agent with the capability to consider the cause-effects. These improvements would narrow the gap between AI and human capabilities and broaden the adoption of RL in real-world applications.Read moreRead less