New fundamental AI is supercharging the search in research

News / Davina Montgomery / March 7, 2024

Key points:

  • ARC Laureate research leads to the creation of AI algorithms that allow scientists to tackle far more complex questions in a fraction of the time.
  • New AI developed at Deakin University’s Applied AI Institute can accelerate the search for solutions in almost any area of discovery.
  • This world-leading development means machine learning can be used to reduce the cost of complex research dramatically.

How AI is accelerating scientific discovery

At Deakin University’s sprawling and tree-covered Waurn Ponds campus in Geelong, a team of fundamental artificial intelligence researchers is quietly expanding the field of what is possible to achieve with AI.

At the helm of this work is Co-Director of Deakin’s Applied AI Institute, Alfred Deakin Professor Svetha Venkatesh. Prof Venkatesh is one of the Top 15 women in AI in the world and Australia’s foremost expert in Bayesian Optimization, a specialised field of AI that employs elegant algorithms to answer questions that are too complex for other forms of machine learning to tackle.

Prof Venkatesh has led her team of researchers to develop a suite of new algorithms that can rapidly accelerate the process of experimentation across almost any field of scientific research, or complex industry research and development.


Professor Svetha Venkatesh


In 2018, Prof Venkatesh was awarded an Australian Research Council (ARC) Laureate, Australia’s most prestigious scientific award, for her world-leading expertise in pattern recognition and what she calls ‘sample-efficient’ AI.

The Laureate seeded a $5 million artificial intelligence research project focused on one key question: can we use machine learning to reduce the number of experiments?

‘That means that you can get to your target faster and cheaper,’ Prof Venkatesh said, adding that the results have been astounding.

‘We’ve been able to reduce the complexity of search, not by 5% or even 50%, but by 20 or 30 times faster than was previously possible.’


Real results, fast

The acceleration of results meant that the team could move far beyond the original scope of the Laureate project, tackling ever more complex machine learning questions:

What happens if the number of control variables that you have increases and it’s very large? There were no algorithms for this.

What happens if you want to transfer the learnings from one experiment to the other? There were few algorithms for this at that time.

How will I work with people who are in some other discipline and apply this and get it to actually work?

A major piece of work conducted in partnership with the Black Dog Institute used the same technology to accelerate trials of personalised mental health treatments for students.


Prof Svetha Venkatesh at work with Prof Sunil Gupta, Prof Truyen Tran, A/Prof Santu Rana, and A/Prof Shannon Ryan
Prof Svetha Venkatesh at work with Prof Sunil Gupta, Prof Truyen Tran, A/Prof Santu Rana, and A/Prof Shannon Ryan


Theorising that this approach could work across a broad range of disciplines, the team worked with leading scientists in multiple fields to see how these new AI methods could accelerate on the ground research projects.

A project with materials scientists and HeiQ Australia Pty Ltd used the AI to cut the design time of new short polymer fibres by 96%.

Working with metallurgists at Deakin’s Institute for Frontier Materials, the composition and processing of new alloys was accelerated more than five times using these AI methods.

Partnering with leading bioprinting experts, Bayesian Optimization AI was employed for the first time in 3D extrustion bioprinting to produce new cell scaffolding.

And with Deakin’s Global Obesity Centre, a World Health Organization Collaborating Centre, AI was used help GPs pinpoint effective ways to encourage patients with obesity to increase their physical activity.

The potential impact for accelerating scientific discovery is huge.

‘As an example, an alloy normally takes two years to make, and we’ve made one here with a team of materials scientists at Deakin within six months. So, it’s a huge amount of savings in time,’ Prof Venkatesh said.


The next questions for sample-efficient AI

One of the questions Prof Venkatesh and her team are looking at now is how to ask the next question.

‘I’m trying to see if we can build machines that can help us ask the next questions and to keep searching in the unknown. Because humans have such uniquely interesting qualities of creativity and curiosity and imagination, and the question is how do we amplify them?

‘I think more and more that will be the direction in the future, where these machines become more like neuro-symbolic AI agents that help you in your regular work.’


This work was made possible by the ARC Australian Laureate Fellowships scheme.

Find the Media Release here


The text of this article is licensed under the Creative Commons Attribution (CC BY) 4.0 International license. We’d love for you to share it, so feel free! Please note that images, videos, graphics and logos are not covered by the CC BY license and may not be used without permission from Deakin University or their respective copyright holder. If you have any questions please contact Thanks for reading! You can find more stories like this in our news section. We ask that Deakin University’s Applied AI Institute, Deakin University and individuals are appropriately credited and that you include links back to this website. Quotes in this article can be extracted for other articles provided individuals are appropriately credited and you include a link back to the article URL.