1Most modeling in systems neuroscience has been descriptive where neural 2 representations, that is, 'receptive fields', have been found by statistically 3 correlating neural activity to sensory input. In the traditional physics approach 4 to modelling, hypotheses are represented by mechanistic models based on the 5 underlying building blocks of the system, and candidate models are validated 6 by comparing with experiments. Until now validation of mechanistic cortical 7 network models has been based on comparison with neuronal spikes, found 8 from the high-frequency part of extracellular electrical potentials. In this 9 * Joint first author † Joint first author ‡ Corresponding author: gaute.einevoll@nmbu.no the signal, the local field potential (LFP), can be used to infer properties of the 11 neuronal network. In particular, we asked the question whether the LFP can 12 be used to accurately estimate synaptic connection weights in the underlying 13 network. We considered the thoroughly analysed Brunel network comprising 14 an excitatory and an inhibitory population of recurrently connected integrate-15 and-fire (LIF) neurons. This model exhibits a high diversity of spiking 16 network dynamics depending on the values of only three synaptic weight 17 parameters. The LFP generated by the network was computed using a hybrid 18 scheme where spikes computed from the point-neuron network were replayed 19on biophysically detailed multicompartmental neurons. We assessed how 20 accurately the three model parameters could be estimated from power spectra 21 of stationary 'background' LFP signals by application of convolutional neural 22 nets (CNNs). All network parameters could be very accurately estimated, 23 suggesting that LFPs indeed can be used for network model validation. 24 Significance statement 25 Most of what we have learned about brain networks in vivo have come from the 26 measurement of spikes (action potentials) recorded by extracellular electrodes.
27The low-frequency part of these signals, the local field potential (LFP), 28 contains unique information about how dendrites in neuronal populations 29 integrate synaptic inputs, but has so far played a lesser role. To investigate 30 whether the LFP can be used to validate network models, we computed LFP 31 signals for a recurrent network model (the Brunel network) for which the 32 3 ground-truth parameters are known. By application of convolutional neural 33 nets (CNNs) we found that the synaptic weights indeed could be accurately 34 estimated from 'background' LFP signals, suggesting a future key role for 35 LFP in development of network models. 36 1 Introduction 37 The traditional physics approach to modeling typically involves four steps: 38 (i) A hypothesis is formulated in terms of a candidate mechanistic mathemat-39 ical model, that is, a model based on interactions between building blocks of 40 the system, (ii) predictions of experimentally measurable quantities are calcu-41 lated from the model, (iii) the predictions are compared with exper...