Where would you drive Machine Learning and Artificial Intelligence (ML/AI) for the next decade?

CSIRO, Australia’s national science agency, is bringing the deep learning revolution to genetics. Improving decision making. Building frameworks for ML-ready data. Predicting change in complex systems. And more.

Future Science Platforms (FSPs) are an investment in science that underpins innovation and that has the potential to help reinvent and create new industries for Australia. FSPs will see us grow the capability of new generation of researchers and allow Australia to attract the best students and experts to work with us on future science.

Many of the challenges facing our society require multidisciplinary solutions which are larger than a single human brain can solve. Machine learning provides the opportunity to solve science challenges using data-driven or model-driven science, increased data interpretation speeds, or increased speed of data analysis. Digital technologies will be one of the key drivers of new industries in coming decades, and to be ready for this change we need to have new approaches to understanding increasingly complex, large and interlinked data sets, and to ensure that these approaches are interpretable, scalable, ethical and trustable.

Machine learning and artificial intelligence (ML/AI) are capabilities that will transform economies and the basis of competition globally, unlock new societal and environmental value and accelerate scientific discovery.

The ML/AI FSP will be the first to work across the whole of CSIRO, building cross-disciplinary projects that apply ML/AI to explore fundamental questions about conceptual and data-driven research applications. The solutions, platforms and people trained through the ML/AI FSP will create a new capability within CSIRO to address core research challenges for the benefit of Australia.

The ML/AI FSP will explore questions such as: how do we use machine learning to augment a scientist’s ability to generate and learn from scientific data? What is the best way to include domain constraints (such as physical laws) and design constraints (such as privacy and fairness) into machine learning models? Where can we exploit genomic information in plant and animal breeding? Why is deep learning so effective in extracting meaningful features? How can we provide explainable AI for decision-making to protect the great barrier reef? Solving these types of challenges will open new vistas of scientific knowledge and positive impact.

The goals of the FSP are:

  • Science – the FSP is an investment that will deliver lasting impact to areas of strategic interest across CSIRO by exploiting and advancing ML/AI research.
  • Technology – we will deliver new ML/AI solutions to age-old problems, novel solutions, and platforms for emerging challenges in a data driven world.
  • People – the platform we create, and the people we train will become a capability that fundamentally changes the way CSIRO undertakes core research challenges.

The ML/AI FSP is designed to bring machine learning to CSIRO’s science. It accommodates areas of expertise defined through a consultative process, and is structured into “activities”, areas which can transform the science undertaken by CSIRO. The FSP seeks to consolidate the applications of ML into science across organisational boundaries.

Our Activates are:

Name Activity
Bioprediction Transforming biological production systems
Constraints MLAI models with design constraints, e.g. scalability, uncertainty propagation, and privacy
Hybrid prediction MLAI with physical process models and constraints for prediction. Robust and efficient approximation of process models using emulators or surrogate models
Object detection Development of a general feature extraction platform, and methods to automate data labelling and synthetic data generation, for image and image like data
Spatiotemporal Predicting change and attributes in complex systems in space and time by integrating diverse data and knowledge
Decisions Decision making and reinforcement learning, active annotation and Bayesian optimisation; verifiable, explainable, ethical ML/AI
Interfaces Descriptive framework for integrating MLAI methods with CSIRO datasets; knowledge sharing platform

Open positions (in partnership with major Australian universities) will be advertised here, and please contact mlaifsp_admin@csiro.au for queries about research partnerships.

Follow us on Twitter @CsiroMlai.