Kavli Affiliate: Kay Tye and Terrence Sejnowski
| Authors: Ben Tsuda, Stefan C Pate, Kay M Tye, Hava T Siegelmann and Terrence J Sejnowski
| Summary:
Neuromodulators are critical controllers of neural states, with dysfunctions linked to various neuropsychiatric disorders. Although many biological aspects of neuromodulation have been studied, the computational principles underlying how neuromodulation of distributed neural populations controls brain states remain unclear. Compared with specific contextual inputs, neuromodulation is a single scalar signal that is broadcast broadly to many neurons. We model the modulation of synaptic weight in a recurrent neural network model and show that neuromodulators can dramatically alter the function of a network, even when highly simplified. We find that under structural constraints like those in brains, this provides a fundamental mechanism that can increase the computational capability and flexibility of a neural network. Diffuse synaptic weight modulation enables storage of multiple memories using a common set of synapses that are able to generate diverse, even diametrically opposed, behaviors. Our findings help explain how neuromodulators “unlock” specific behaviors by creating task-specific hyperchannels in the space of neural activities and motivate more flexible, compact and capable machine learning architectures. Significance Neuromodulation through the release of molecules like serotonin and dopamine provides a control mechanism that allows brains to shift into distinct behavioral modes. We use an artificial neural network model to show how the action of neuromodulatory molecules acting as a broadcast signal on synaptic connections enables flexible and smooth behavioral shifting. We find that individual networks exhibit idiosyncratic sensitivities to neuromodulation under identical training conditions, highlighting a principle underlying behavioral variability. Network sensitivity is tied to the geometry of network activity dynamics, which provides an explanation for why different types of neuromodulation (molecular vs direct current modulation) have different behavioral effects. Our work suggests experiments to test biological hypotheses and paths forward in the development of flexible artificial intelligence systems.