The first two activities span the full spectrum of basic research to practical application: MLAI applications in the life sciences (Bioprediction) and Spatiotemporal problems. Next we have two method driven activities. The first (Context) to capture desirable design constraints such as privacy and fairness, into current machine learning algorithms along with socio-technical platforms for specifying MLAI tasks. The second (Decisions) to take us from machine learning predictions to decision making and actions. In science we often have very specific domain expertise, which we want to exploit in a hybrid fashion with MLAI predictions. We call this activity Hybrid prediction. One of the major successes of MLAI is in the field of computer vision, which we will use to assist scientists to identify objects and extract features from raw data. The activity leveraging this for CSIRO science is called Object detection.
MLAI models with design constraints, e.g. scalability, uncertainty propagation, and privacy.
Decision making and reinforcement learning, active annotation and Bayesian optimisation; verifiable, explainable, ethical ML/AI.
Building predictive models that are a hybrid of data driven statistical machine learning and domain knowledge. Through Hybrid Prediction, we […]
Development of a general feature extraction platform, and methods to automate data labelling and synthetic data generation, for image and image like data.