Rethinking Topological Persistence (2024-2028)

This project aims to address the lack of transferability and uncertainty-awareness in AI models. Despite their success, AI models are met with bias and uncertainty when deployed in the real world. As a result, they are rarely used in high-risk industries like cybersecurity or transport. This project expects to build uncertainty- awareness into models by teaching them to return UNKNOWN when they encounter a previously unseen thing, instead of misclassifying it. Further, the evaluation methods to be developed will not rely on access to test data, allowing cost- effective, private, and safe AI for high- stakes decision support. The outcomes will benefit Australia by accelerating economic investment and fostering greater social acceptance of AI.