Neural Computation Laboratory
The Neural Computation Laboratory is directed by Stefano Panzeri and aims at understanding how circuits of neurons in the brain exchange and transmit information and contribute to sensation and behavior.
The laboratory addresses this issue by developing advanced statistical tools for the analysis of simultaneous recordings of neural activity from multiple locations, by applying these tools to empirical data to understand how neurons encode and transmit information, and by developing biophysically plausible models of neural circuit dynamics that explain the empirical findings.
Mathematical methods for cracking the neural code using recordings of neural activity at different locations and spatio-temporal scales.
We produced original mathematical methods, based on the mathematical principles of communication theory, for the analysis of neural data recorded simultaneously from multiple sites, even at different spatial and temporal scales. These methods determine which spatial and temporal features of neural activity encode sensory stimuli, how correlations between these features affect information processing, and if and how their information is used to produce useful behavior.
The role of the temporal structure of neural responses in sensory information coding.
In collaboration with several neurophysiological laboratories, we apply the mathematical methods developed by us to provide a number of novel insights about how the cerebral cortex uses the timing of neural activity to encode information about the natural environment in visual, auditory and somatosensory neural systems. In recent work, we have demonstrated the role of the millisecond precise spike timing in encoding sensory information, the role of network-level oscillations (spanning several octaves of frequency) in encoding and transmitting sensory information, and the mechanisms for multiplexing information in spike times and oscillations at multiple concurrent temporal scales.
Recurrent neural network models of how cortical circuits encode information.
We developed biophysically plausible yet analytically tractable models of recurrent spiking neural that can be quantitatively fit to data to establish how activity of networks of excitatory and inhibitory neurons encode the time and frequency structure of their inputs, and how changes in excitation and inhibition or neuromodulation affect how cortical circuits process and transmit information.
Theoretical research in the lab is mostly carried out in the Center for Neuroscience and Cognitive Systems at IIT Rovereto. The laboratory is equipped with state of the art computational facilities. We currently feature a cluster with a total of more than 700 processors high-end Intel Xeon processors with a total of more than 2TB of RAM memory and 60 TB of disk, hosted in 16 servers.