Researchers from the University of Liège have developed an algorithm that’s based on a biological mechanism called neuromodulation.
The field of Artificial Intelligence has enjoyed immense progress in recent times. But, despite these advancements, we’re still very far from human intelligence.
Yes, current AI training techniques can allow computer agents to perform some tasks better than humans.
However, these machines are often trained for a specific task. So, they produce disappointing results in conditions that are slightly different from what they experienced during training.
Humans, on the other hand, are capable of adapting to new situations. We can use the skill acquired throughout our life to adjust to unforeseen conditions.
For example, a child who has learned to walk indoor can quickly learn to walk outdoor as well.
In this context, we can associate “learning to walk” with synaptic plasticity, which modifies the connection between neurons. Meanwhile, the rapid adaptation of walking skills learned indoor to those needed to walk outdoor is linked with neuromodulation.
Neuromodulation and Synaptic Plasticity
Neuromodulation is the mechanism that dynamically controls how neurons respond to external stimuli in a context-dependent manner. It allows us to adapt quickly to unforeseen problems.
Meanwhile, synaptic plasticity is the process by which specific patterns of synaptic activity result in changes in synaptic strength. Researchers believe it contributes to learning and memory.
Current advances in AI are partly due to synaptic plasticity. But, scientists have never figured out a way to introduce neuromodulation to AI, until now.
Co-author of the study and AI specialist, Damien Ernst said:
“The novelty of this research is that, for the first time, cognitive mechanisms identified in neuroscience are finding algorithmic applications in a multitasking context.”
Other researchers that participated in the study include Nicolas Vecoven, Antoine Wehenkel, and Guillaume Drion. In their paper in PLOS ONE, the team described how they developed an algorithm based on neuromodulation.
Introducing Neuromodulation to Artificial Neural Networks
The ULiège team developed a unique artificial neural network and introduced an interaction between two sub-networks.
The first sub-network considers all the contextual information on a task that it needs to solve. Then, it uses the data to neuro module the second sub-network in a way that’s reminiscent of the brain‘s chemical neuromodulators.
As a result, the second sub-network — which determines the action an intelligent agent needs to perform — can be adapted very quickly to new conditions. Simply put, it allows the agent to solve new tasks efficiently.
The team has successfully tested the innovative architecture on classes of navigation problems that require adaptation.
They trained AI agents to move towards a specific target while avoiding obstacles on the way. However, the agents were able to adapt to situations in which a variable wind direction disrupted their movement.
“This research opens perspectives in the exploitation in AI of neuromodulation, a key mechanism in the functioning of the human brain,” says Ernst
Comments (0)
Most Recent