The response from many neurons needs to be pooled together in order to reconstruct the sensory input (the hidden cause) from the firing rates of neurons. There is an optimal way to combine the neurons from the population in order to reconstruct the direction of the moving stimuli (Jazayeri and Movshon 2006). An optimal combination is obtained when the weights of each neuron are adjusted with respect to their tuning function.
However, it was currently unknown of these optimal gain could arise in the brain. In this theoretical paper, Habenschuss and colleagues suggest that spike-timing dependent plasticity, an elementary plasticity rule, can lead to optimal gains.
With this very simple model, the authors demonstrate that this plasticity rule leads to near-optimal decoding performance. They show that they were able to reconstruct the sensory input from the signal of the noisy sensory neurons as accurately as a Bayes optimal decoder would do. In addition, they show that the optimality was preserved when sensory neurons were removed or added and when tuning functions were modified. These two phenomena are known to occur frequently in the brain. Interestingly, they also show that more frequent inputs affected the readout population more strongly (their weights were larger). Such a phenomena has also been observed indirectly.
The main message of this article is that spike-timing dependent plasticity account for optimal decoding of sensory neurons. It would be extremely interesting to study whether this same rule could explain other aspects of Bayesian inferences such as the combination of sensory inputs from different sensory modalities with respect to their uncertainty. It might also be interesting to see how uncertainty about the decoded direction could be represented in such a network.
Habenschuss, S., Puhr, H. & Maass, W. Emergence of Optimal Decoding of Population Codes Through STDP. Neural computation 25, 1371–407 (2013).
Albright TD (1984) Direction and orientation selectivity of neurons in visual area MT of the macaque. Journal of neurophysiology 52:1106–1130.