{"links":{"self":"http://dataportal.arc.gov.au/NCGP/API/grants/DE260101202"},"data":{"type":"grant-details","id":"DE260101202","attributes":{"code":"DE260101202","administering-organisation":"Adelaide University","announcement-administering-organisation":"The University of Adelaide","scheme-name":"Discovery Early Career Researcher Award","grant-status":"Active","funding-commencement-year":2026,"years-funded":3,"project-start-date":"2026-12-01","anticipated-end-date":"2029-11-30","grant-summary":"Efficient Generative Models with Enhanced Perception and Human Alignment. This DECRA project aims to develop efficient training strategies for generative models (GMs) with enhanced perception and human alignment, advancing generative AI toward interactive AI. Current GMs suffer from inefficiency, limited perceptual awareness, and challenges in human alignment and controllability. This project expects to advance GMs with interactive self-improvement that dynamically learn from inherent knowledge, integrate contextual perception, align seamlessly with human guidance, and incorporate self-evaluation. The outcomes will establish robust GMs for real-world applications like climate prediction, personalized healthcare AI and manufacturing, benefiting Australia’s society, economy, and research capacity in generative AI.","funding-current":509811.00,"funding-at-announcement":505678,"investigators-current":[{"title":"Dr","firstName":"Xinyu","familyName":"Zhang","roleName":"Discovery Early Career Researcher Award","roleCode":"DECRA","isFellowship":true,"orcidIdentifier":"0000-0002-2999-3291 "}],"investigators-at-announcement":[{"title":"Dr","firstName":"Xinyu","familyName":"Zhang","roleName":"Discovery Early Career Researcher Award","roleCode":"DECRA","isFellowship":true,"orcidIdentifier":"0000-0002-2999-3291 "}],"organisations-current":[{"organisationName":"Adelaide University","roleName":"Administering Organisation","state":"SA"}],"organisations-at-announcement":[{"organisationName":"The University of Adelaide","roleName":"Administering Organisation","state":"SA"}],"field-of-research":[{"isPrimary":false,"code":"460304","name":"Computer Vision","type":"FOR20"},{"isPrimary":true,"code":"4605","name":"Data Management and Data Science","type":"FOR20"},{"isPrimary":false,"code":"460502","name":"Data Mining and Knowledge Discovery","type":"FOR20"},{"isPrimary":false,"code":"461103","name":"Deep Learning","type":"FOR20"}],"socio-economic-objective":[{"code":"220403","name":"Artificial Intelligence","type":"SEO20"},{"code":"280115","name":"Expanding Knowledge In the Information and Computing Sciences","type":"SEO20"}],"international-collaboration":["Singapore","United States of America"],"lief-register":[],"achievement-summary":null,"national-interest-test-statement":"This project aims to develop advanced generative artificial intelligence (AI) that can create reliable synthetic data to address data scarcity challenges. Current AI systems require extensive training resources and often produce inaccurate or unhelpful content due to a lack of contextual understanding. In contrast, the proposed system reduces the need for large-scale training while improving its ability to interpret real-world situations and align with human expectations, ensuring more reliable and controllable outputs. The synthetic data generated by this system can enhance decision-making models across diverse real-world scenarios. For example, in autonomous driving, it can generate realistic data for rare weather conditions like heavy rain or dense fog to make self-driving safer. In medical diagnosis, it can create synthetic radiological images of rare cancer cases to enhance diagnosis accuracy. In manufacturing, it can support anomaly detection by providing more examples of defective and anomaly products, improving quality control. This research will also provide strong benefits to Australians economy, environment, and society. For example, aligning outputs with human values can help mitigate the risk of inappropriate or harmful content, thus strengthening cybersecurity. The outcomes of this project will be promoted via public engagement (workshops, seminars and public talks), online platforms (Twitter and GitHub), open access publications, and industry collaborations."}}}