Search

The brain prepares for learning even before interacting with the environment, by refining and optimizing its structures through spontaneous neural activity that resembles random noise. However, the mechanism of such a process has yet to be thoroughly understood, and it is unclear whether this process can benefit the algorithm of machine learning. Here, we study this issue using a neural network with a feedback alignment algorithm, demonstrating that pretraining neural networks with random noise increases the learning efficiency as well as generalization abilities without weight transport. First, we found that random noise training modifies forward weights to match backward synaptic feedback, which is necessary for teaching errors by feedback alignment. As a result, a network with pre-aligned weights learns notably faster than a network without random noise training, even reaching a convergence speed comparable to that of a backpropagation algorithm. Sequential training with both random noise and data brings weights closer to synaptic feedback than training solely with data, enabling more precise credit assignment and faster learning. We also found that each readout probability approaches the chance level and that the effective dimensionality of weights decreases in a network pretrained with random noise. This pre-regularization allows the network to learn simple solutions of a low rank, reducing the generalization loss during subsequent training. This also enables the network robustly to generalize a novel, out-of-distribution dataset. Lastly, we confirmed that random noise pretraining reduces the amount of meta-loss, enhancing the network ability to adapt to various tasks. Overall, our results suggest that random noise training with feedback alignment offers a straightforward yet effective method of pretraining that facilitates quick and reliable learning without weight transport.

Social deficit is a major feature of neuropsychiatric disorders, including autism spectrum disorders, schizophrenia, and attention-deficit/hyperactivity disorder, but its neural mechanisms remain unclear. Here, we examined neuronal discharge characteristics in the medial prefrontal cortex (mPFC) of IRSp53/Baiap2-mutant mice, which show social deficits, during social approach. We found a decrease in the proportion of IRSp53-mutant excitatory mPFC neurons encoding social information, but not that encoding non-social information. In addition, the firing activity of IRSp53-mutant neurons was less differential between social and non-social targets. IRSp53-mutant excitatory mPFC neurons displayed an increase in baseline neuronal firing, but decreases in the variability and dynamic range of firing as well as burst firing during social and non-social target approaches compared to wild-type controls. Treatment of memantine, an NMDA receptor antagonist that rescues social deficit in IRSp53-mutant mice, alleviates the reduced burst firing of IRSp53-mutant pyramidal mPFC neurons. These results suggest that suppressed neuronal activity dynamics and burst firing may underlie impaired cortical encoding of social information and social behaviors in IRSp53-mutant mice.

NMDA receptor (NMDAR) and GABA neuronal dysfunctions are observed in animal models of autism spectrum disorders, but how these dysfunctions impair social cognition and behavior remains unclear. We report here that NMDARs in cortical parvalbumin (Pv)-positive interneurons cooperate with gap junctions to promote high-frequency (>80 Hz) Pv neuronal burst firing and social cognition. Shank2-/- mice, displaying improved sociability upon NMDAR activation, show impaired cortical social representation and inhibitory neuronal burst firing. Cortical Shank2-/- Pv neurons show decreased NMDAR activity, which suppresses the cooperation between NMDARs and gap junctions (GJs) for normal burst firing. Shank2-/- Pv neurons show compensatory increases in GJ activity that are not sufficient for social rescue. However, optogenetic boosting of Pv neuronal bursts, requiring GJs, rescues cortical social cognition in Shank2-/- mice, similar to the NMDAR-dependent social rescue. Therefore, NMDARs and gap junctions cooperate to promote cortical Pv neuronal bursts and social cognition.