Oscillations and waves, the lingua franca of computations in the cerebral cortex

Wolf Singer (Max Planck Institute for Brain research, Ernst-Strüngmann-Institute, Frankfurt am Main)

The astonishing performance of deep neuronal networks (DNNs) and large language models (LMNs) suggests that these artificial systems capture essential properties of natural intelligent systems. However, the computational architectures of artificial and natural systems differ in the way they process temporally structured information. In artificial systems the temporal dimension needs to be simulated, in natural systems it it is an emergent property of their dynamics. Results from simulations studies suggest, that the cerebral cortex uses the dynamics of recurrent networks, whose nodes function as damped harmonic oscillators, for computation. By exploiting the unique properties of coupled oscillators these harmonic oscillator recurrent networks (HORNs) acquire surprising efficiency in bench-mark tasks for pattern recognition with respect to learning speed, noise tolerance and sparsity of parameters. Moreover, they reproduce numerous, experimentally confirmed properties of the cerebral cortex. It is proposed that oscillations, travelling waves and the resulting interference patterns observed in experiments serve basic computational operations rather than being inevitable epiphenomena of complex neuronal interactions.