Home Research Staff Projects


Dr. Urs Braun

Boehringer Ingelheim Stiftung : Recurrent neural networks as preclinical models of cognitive dysfunction. 10/2021-03/2023.

Recurrent neural networks (RNN) are a class of artificial neural networks that can be trained with empirical behavioral data to emulate in silico how the brain performs complex cognitive tasks. This training is achieved by adapting a complex topological structure that is optimized to perform algorithmic computations through a nonlinear input-output mapping. The resulting properties of RNN can be analyzed with tools from network neuroscience to gain insights into the neural mechanisms of high-order cognition, and have shown remarkable similarities to the topological properties of the living brain. The proposed project aims to establish a framework to enable the use of RNN as preclinical models of cognition in health and disease. Capitalizing on schizophrenia as an example of cognitive dysfunction, the project will examine the organizational properties of RNN trained on aggregated cognitive performance data derived from a battery of cognitive tests in controls and patients with schizophrenia. We expect that RNN trained on human healthy data will show network properties that support a balance between segregated and integrated information processing, while RNN trained on schizophrenia data will show a shifted balance towards integrated processing. To further validate the utility of RNN as cognitive models and demonstrate their sensitivity to inter-individual differences, we will collect highly sampled longitudinal cognitive test data from 5 healthy subjects over the course of one year, and train individual RNN on those test data. We expect individual models to show topological variation that explains individual variation in independent cognitive measures.

Zentralinstitut für Seelische Gesundheit (ZI) - https://www.zi-mannheim.de