What if…? Predicting the future is of fundamental importance for our interaction with the world, for learning and planning. There is evidence that this also applies to various areas of our brain: nerve cells can anticipate sensory impulses and become active in advance. This is the subject of a paper recently published by researchers from the Ernst Strüngmann Institute (ESI) for Neuroscience in the renowned scientific journal Nature Communications. In their article Sequence Anticipation and Spike-Timing-Dependent Plasticity, Matteo Saponati and Martin Vinck discuss how our brain can predict future events, an ability that is crucial for intelligent behavior.

What principles are used to evaluate and classify inputs?

Nerve cells in the brain, also known as neurons, receive a variety of information (inputs) through synapses that connect to them. Organisms rely on anticipating future inputs in order to interact with the world and organize their behaviour accordingly. In their current work, Matteo Saponati and Martin Vinck ask: What principles are used to evaluate and classify inputs? Which learning rules help neurons to predict inputs? What role does learning (synaptic plasticity) play in predicting future events?

They conclude that neurons rate synaptic inputs with inverse proportionality to their predictability: Neurons therefore treat the information that synapses pass on to them depending on how easy it is to predict what will happen next. If a connection is good at predicting what will happen next, it is strengthened. In short, neurons strengthen the inputs that predict the occurrence of subsequent inputs and suppress the predicted inputs themselves. Following the same principle, neurons also learn to predict high-dimensional patterns over short and long temporal sequences: Neurons fire strongly for the first inputs, which are still unpredictable, and then suppress the subsequent, predictable inputs.

Neurons learn to anticipate by developing a model of how different inputs relate to each other over time. At the same time, this learning mechanism enables sequences of events to be learned over long periods of time and activity patterns to be adapted accordingly.

A learning rule that can explain several phenomena

The researchers also suggest that this learning rule can explain several phenomena already observed experimentally in the brain, in particular spike-timing-dependent plasticity (STDP), a process in which the timing of neuron firing has a significant impact. Prediction appears to play a fundamental role here as to how individual neurons learn and adapt because it can influence the timing and organization of neuronal activity.

The two ESI scientists and their team hope that the findings of their study will help researchers in the machine learning and computational neurosciences for building novel biologically-inspired training algorithms. Researchers in the the field of neuromorphic computing could also benefit from their results by considering and applying this model in the design of new chips.

Original Publication
Saponati M, Vinck M (2023). Sequence anticipation and spike-timing-dependent plasticity emerge from a predictive learning rule. Nature Communications, 14, 4985. https://doi.org/10.1038/s41467-023-40651-w