Computational Modeling

Computational Modeling

Our sensory systems perform the complex task of encoding and allowing us to perceive our dynamic and fast-paced environment. Although the sensory percepts are stable and consistent, the neural activity that encodes this information is variable and noisy. We are fundamentally interesting in understanding how a reliable percept can be decoded from this neural activity. We utilize computational models to identify these complex system dynamics and to generate  statistical models to predict behavioral outputs from our recorded electrophysiological data.

Neural System Identification & Phenomenological Modeling

System identification has phenomonological modela long history of application in sensory neuroscience for creating nonlinear dynamical models. In vivo experimental models are typically limited in the amount of data that can be realistically collected during an experimental session, often precluding the high-order modeling that is necessary to adequately capture the complexity of the system. However, through a combination of nonparametric modeling and empirical observations of the system, it is possible to constrain the model subspace, greatly reducing the number of parameters needed, while only minimally restricting the generality of the model.

Systematic probing of the input–output relationship enables the development of  nonlinear phenomenological models based upon experimental observations that is highly predictive of the cortical response to patterns of stimulation. Through both nonparametric and parametric system identification techniques, we have developed models of the nonlinear system dynamics in the thalamocortical circuit using our in vivo electrophysiological data as a foundation. More generally, the nonlinear dynamical models developed in our lab can inform future encoding schemes that map sensory signals to patterns of microstimulation for sensory prosthesis implementation.

Select Publications

D. C. Millard, Q. Wang, C. A. Gollnick, and G. B. Stanley. System identification of the nonlinear dynamics in the thalamocortical circuit in response to patterned thalamic microstimulation in-vivo, J Neural Eng., Dec;10(6):066011, 2013. PDF

Single Neuron and Network Modeling of the Thalamic Circuit

SensorTC_fig2y systems serve the purpose of allowing us to extract perceptually relevant features from the environment. Although there are certainly examples of sensory features whose coding originates in the sensory periphery (e.g. auditory frequency, visual color, etc.), the more intriguing and less well understood phenomena involve the emergence of feature selectivity in more central brain structures that do not just inherit the selectivity from the periphery. We can use simple computation models of neurons to determine and probe the biophysical mechanisms and encoding strategies behind feature selectivity in the thalamo-cortical circuit. In particular, we have developed primarily simple neural models, such as integrate-and-fire (I&F), integrate-and-fire-and-burst( I&F-B), and Izhikevich Thalamo-cortical neurons to model and predict neural phenomena observed in our in vivo electrophysiological data.

Select Publications:

S. Boloori, R. A. Jenks, Gaelle Desbordes, and G. B. Stanley. Encoding and decoding cortical representations of tactile features in the vibrissa system, J. Neurosci., 30(30):9990-10005, 2010. PDF

R. Ollerenshaw, H. J. V. Zheng, Q. Wang, and G. B. Stanley, The adaptive trade-off between detection and discrimination in cortical representations and behavior,Neuron., Mar 5;81(5):1152-64, 2014. PDF

T. Kelly, J. Kremkow, J. Jin, Y. Wang, Q. Wang, J. M. Alonso, G. B. Stanley. The Role of Thalamic Population Synchrony In the Emergence of Cortical Feature Selectivity,PLoS Comput Biol., Jan;10(1):e1003418, 2014. PDF

A. Lesica, J. Z. Jin, C. Weng, C. I. Yeh, D. A. Butts, G. B. Stanley, and J. M. Alonso. Adaptation to stimulus contrast and correlations during natural visual stimulation, Neuron, 55:479-491, 2007. PDF

Statistical Modeling

One of our objectives is to use knowledge of the stimulus and ongoing neural information to predict the neural and behavioral response.  At the core of this perspective, is expressing the probability of observing a neuronal response R given the presentation of the stimulus S, or p(R|S), which is generally referred to as an encoding model. How much does the observation of the neuronal response reduce my uncertainty about the stimulus S? Using our electrophysiological data from our in vivo experiments, we can develop encoding models to predict neural responses to ongoing activity, using the statistics of the neural firing rate. Additionally, we can develop statistical models to give insight into the complex dynamics of neural systems and the neural code. For example, by providing a white noise stimulus to sensory responsive neurons, we can perform spike triggered averaging to extract linear filters that can be predictive of the neurons tuning curve. By developing multiple linear filters from a single population of neurons ,we can extract commonalities across the region to predict the populations response to ongoing stimuli. By using statistical information about the neuron, we can begin to predict neural response to complex stimuli.

Select Publications

Lesica, N. A., C. Weng, J. Jin, C. Yeh, J. M. Alonso, and G. B. Stanley. Dynamic encoding of natural luminance patterns by LGN bursts, PLOS Biology, 4(7), e209, 2006. PDF

Stanley, G. B., J. Z. Jin, Y. Wang, G. Desbordes, M. J. Black, J. M. Alonso. Visual orientation and direction selectivity through thalamic synchrony, J. Neurosci., 32, 9073-9088, 2012. PDF

Desbordes, G., J. Z. Jin, J. M. Alonso, G. B. Stanley. Modulation of temporal precision in thalamic population responses to natural visual stimuli, Frontiers in Systems Neuroscience, 4:151, 2010. PDF