My main research focus challenges the traditional view of biological neurons as simple spatial pattern recognizers, instead revealing them as sophisticated devices capable of highly complex spatio-temporal pattern recognition. 

 

We focus on understanding how single neurons and small networks learn, represent and process information, with particular emphasis on the computational roles of various biological features, such as dendritic morphology and nonlinearities, multiple synaptic contacts, firing rate adaption, the specific structure of cortical/hippocampal/cerebellar microcircuits, etc. 

We are curious about how these various biological “details” aid and contribute to various forms of computation. 

 

Our past work has demonstrated that the input-output relationship of a single Layer 5 cortical neuron can be accurately modeled using a deep neural network with 5-8 layers, highlighting the remarkable computational complexity of individual neurons and the critical role of NMDA ion channel and dendritic morphology for this complexity. 

Additionally, we investigated how the combination of two seemingly unrelated biological features such as multiple synaptic contacts and dendritic filtering enable neurons to generate precisely timed outputs in response to specific input patterns with increased capacity compared to simplified neurons without these biological features. 

 

Our current research focus is centered around how networks of biophysical spiking neurons can learn to perform complex tasks and how they represent information via spike timing. We utilize deep learning as a key research tool that enables us to tackle questions that were previously impossible to address. We open source all of our code and data so that any researcher around the world can access it and we strive very hard to make them easily accessible by providing thorough documentation.