UNIGE researchers unveil the role played by synaptic feedback systems in the learning processes of the cerebral cortex. A valuable discovery for the development of artificial intelligence.
As in other animals, the huge learning capacity of the human being allows him to apprehend new sensory information and thus master new skills or adapt to a changing environment. But many of the learning mechanisms remain poorly understood. One of the biggest challenges in systems neuroscience is explaining how synaptic connections change to accompany behavioral changes. Neuroscientists at the University of Geneva (UNIGE) had already shown that synaptic learning mechanisms in the cerebral cortex depend on a feedback loop with deep brain regions. They have now shown precisely how this feedback allows synaptic enhancement by activating and deactivating particular inhibitory neurons. This research, to be published in the journlal Neuron, is an important milestone for our understanding of the mechanisms of learning. It can also open new avenues for the design of computerized learning systems and artificial intelligence.
The cortex – the outer and largest area of the brain – is important for higher cognitive functions, complex behaviors, perception, and learning. As soon as a sensory stimulus occurs, the cortex processes and filters the information before returning the most relevant aspects to other areas of the brain. Some of them in turn refer information to the cortex. These loops, called feedback systems, are essential for the functioning of cortical networks and their adaptation to new sensory information. “When learning perceptually (or improving the ability to respond to a sensory stimulus), neural circuits must first assess the importance of sensory information received and then decide how it will be treated. The feedback systems confirm to some extent that the synapses responsible for transmitting information to other brain regions have done so correctly,”says Anthony Holtmaat, Professor of Fundamental Neuroscience at the Faculty of Medicine at UNIGE, who is leading this study.
Mouse whiskers used to shed light on the role of feedback systems
Mouse whiskers, specialized in tactile detection, play a key role in the animal’s ability to understand its direct environment. The part of the cortex that processes the sensory information of whiskers continually optimizes its synapses to incorporate new aspects of the animal’s tactile environment. It is therefore an interesting model for understanding the role of feedback systems in synaptic learning mechanisms.
UNIGE scientists isolated a whisker-related feedback circuit and used electrodes to measure the electrical activity of neurons in the cortex. They then simulated a sensory input by stimulating a specific part of the cortex that processes this information and controlled by a light beam the feedback circuit. “This ex vivo model allowed us to control feedback independently of sensory stimulation, which is impossible to do in vivo, but it was critical to disconnect this feedback stimulation to understand how the interaction between the two leads to synaptic reinforcement,” adds Anthony Holtmaat.
A open door for information
The team found that the two components, when triggered separately, activate a wide range of neurons. However, when they are activated simultaneously, some neurons reduce their activity. “It is interesting to note that inhibited neurons when sensory stimulation and feedback occur simultaneously usually inhibit neurons important to perception, so-called inhibition of inhibition or disinhibition,” says researcher Leena Williams. at the Faculty of Medicine of UNIGE and first author of this study. “These neurons act as if they were opening a normally closed door for incoming information, which helps strengthen the synapses that process the primary sensory information. Our study has helped to identify how feedback optimizes synaptic connections to better interpret future information,” she adds.
Now that they have accurately identified the neurons involved in this mechanism, the researchers will test their results and verify whether the inhibitory neurons behave as expected when a mouse needs to integrate new sensory information or when it discovers new aspects in its tactile environment.
A boost for artificial intelligence
How do brain circuits optimize? How can a system learn on its own by analyzing its own activity? Relevant for the understanding of animal learning, these questions are also at the heart of artificial intelligence and machine learning programs that some specialists are trying to emulate brain circuits to build intelligent systems.
The results obtained by UNIGE researchers could be relevant for autonomous learning, a branch of machine learning that designs circuits capable of self-organizing and optimizing the processing of new information. An additional step towards creating even more effective voice and facial recognition programs.