Olfactory network dynamics and the coding of multidimensional signals

Key Points The brain faces many complex problems when dealing with odorant signals. Odours are multidimensional objects that we usually experience as unitary percepts. They are also noisy and variable, but we can classify and identify them well. The olfactory system therefore solves complicated pattern-learning and pattern-recognition problems. I propose that part of the solution relies on a particular architecture that imposes a dynamic format on odour codes. According to this hypothesis, the olfactory system actively creates a large coding space in which to place odour representations and simultaneously optimize their distribution within it. This process uses both oscillatory and non-periodic dynamic processes that serve complementary roles: slow non-periodic processes allow decorrelation (that is, the reduction of the overlap between odour representations); fast oscillations allow sparsening (that is, a reduction in the size of the coding assemblies) and feature binding (that is, the representation of multiple and co-occurring features by the spikes of single neurons). The prominent role of oscillatory synchronization in the process of sparsening is reviewed. Briefly, sparsening is achieved through a process that involves periodic input, coincidence detection, fan-in and fan-out connection patterns, and delayed feedforward inhibition. These mechanisms together lead to the appearance of rare but highly selective neuronal responses, which synthesize specific combinations of input features. The coding aspects, advantages, disadvantages and possible uses of these interlocked and dynamic integrative phenomena are discussed in the context of olfaction and other systems in which complex sensory objects must be represented, learned and recognized. The brain faces many complex problems when dealing with odorant signals. Odours are multidimensional objects, which we usually experience as unitary percepts. They are also noisy and variable, but we can classify and identify them well. This means that the olfactory system must solve complicated pattern-learning and pattern-recognition problems. I propose that part of the solution relies on a particular architecture that imposes a dynamic format on odour codes. According to this hypothesis, the olfactory system actively creates a large coding space in which to place odour representations and simultaneously optimizes their distribution within it. This process uses both oscillatory and non-periodic dynamic processes wit