ITU researcher secures ERC Consolidator grant to develop rapid learning abilities in AI
Professor Sebastian Risi from the IT University’s Digital Design department has received 1.9 million Euros from the European Research Council to develop AI that is capable of adapting to and learning from unforeseen events in a bid to unleash more of the enormous potential of autonomous machines.
If machines were able to employ the type of rapid learning that characterizes human intelligence, we would be able to automatize a whole range of processes that would impact all parts of society. From machines conducting search-and-rescue missions in difficult environments to something as seemingly simple as emptying a dishwasher without breaking a plate. What these disparate processes have in common is a need for a type of general intelligence which machines have yet to develop.
But that is exactly what Professor Sebastian Risi of the IT University sets out to do with the research project, GROW-AI: Growing Machines Capable of Rapid Learning in Unknown Environments. The researcher has just been awarded a prestigious Consolidator grant from the European Research Council (ERC) to explore the possibilities of creating machines with a more general intelligence, allowing rapid adaption in unknown situations.
Specifically, the researcher will be looking at how biological intelligence is developed in order to optimize artificial neural networks. In stark contrast to current neural networks, whose architectures are designed by human experts and whose large number of parameters are optimized directly, evolution does not operate directly on the parameters of biological nervous systems. Instead, these nervous systems are grown and self-organize through a much smaller genetic program that produces rich behavioral capabilities and the ability to rapidly learn right from birth.
Neuroscience suggests this "genomic bottleneck" is an important regularizing constraint, allowing animals to generalize to new situations. However, currently there does not exist a solution to creating similar systems artificially.
To address the issue, Sebastian Risi and his team will teach neural networks genomic bottleneck algorithms instead of manually designing them and by doing so exploit recent advances in memory-augmented deep neural networks that can learn complex algorithms. In addition, the team will work towards co-optimizing task generators that provide neural networks with the most effective learning environments. Taking inspiration from the fields of artificial life, neurobiology, and machine learning, the researcher proposes investigating if algorithmic growth is needed to understand and create intelligence.
“If successful, this project will greatly improve the autonomy of machines and significantly increase the range of real-world tasks they can solve,” says Sebastian Risi.
“While scaling current neural network approaches to larger and larger models has produced truly ground-breaking results, I’m in doubt that this method alone will get us closer to the goal of a more general AI. With this project I argue for a different paradigm that should allow us to gain a deeper understanding of some ingredients that might be important to create truly intelligent machines.”
The ERC, set up by the European Union in 2007, is the premier European funding organisation for excellent frontier research. It funds creative researchers of any nationality and age, to run projects based across Europe Theis Duelund Jensen, Press Officer, tel: 2555 0447, email: email@example.com