| Zuohua Zhang (Graduate Student ) |
| UR > CS > Research > Vision > People > Zuohua Zhang |
![]() | Contact:Zuohua ZhangDepartment of Computer Science University of Rochester Rochester, NY 14627 email: zzhang@cs.rochester.edu home: http://www.cs.rochester.edu/~zzhang vision: http://www.rochester.edu/research/vision/people/Zouhua_Zhang |
About:Zuohua is in her sixth year of the Computer Science Ph. D. program at the University of Rochester. She is formulating a biologically satisfying model of neural signaling via spikes. | |
Research Interests:A challenge in systems neuroscience is a satisfactory model of neural signaling. From rate coding, which simply uses a spike count, to temporal coding models, which explore temporal structures of spike trains, models of neural signaling have been challenged by the fact that neurons fire highly irregularly, not only during the presentation of sustained stimulus, but also across repeated trials. The variability of neuronal responses is quite contrary to one might expect from those neural coding models. Typically it is treated as caused by noise. Extensive averaging across ensembles of neurons, or across large number of trials, or across time intervals much larger than would seem to be behaviorally relevant, have been used to average away noise for robust estimations of neuronal messages. However, such extensive averaging is not suited for a highly efficient and economical system like the brain, which is the product of millions of years of evolution. Not only firing is metabolically expensive, being such a highly evolved and efficient system, it seems very unlikely that the brain can operate in such a noisy regime with such remarkable efficiency and complexity. Therefore, irregularity of neuronal responses may not be induced by noise, but rather, comprise signals of computation. This ``signal rather than noise'' view is not compatible with either rate coding or temporal coding models. Challenged with the discrepancy between theory and data, we take a fresh view on the subject beginning with the proposal that the randomness associated with neuron outputs is almost certain to have a purpose. In particular, we model neurons as probabilistic devices which are not only computing probabilities but also firing probabilistically to signal their computations. Consistent with neurons coding as ensembles observations and the observations of topological map, signaling of probabilities are done by cells with similar receptive fields firing in synchrony to achieve fast communication. Our proposal of neurons use distributed synchronous spikes to communicate probabilistically not only accounts for the irregularity of spike trains, but also provides the advantage of robust computation. Furthermore, the involvements of probabilistic firing and distributed coding explicate how synchronous firing can appear as a rate code, acknowledging vast amount of data supporting rate code assumption. Any neural signaling model would have to support cortical computation in a biological realistic fashion. Going beyond simply addressing the role of spikes in cortical cells' communication, we show that, our distributed synchrony model can be implemented in a predictive coding framework and be used to learn structures in the natural environment. Trained with patches from natural images, our model V1 cells develop localized and oriented receptive fields, consistent with simple cell properties of V1. Unlike most cortical computation models, our predictive coding model makes the use of single spikes, instead of abstract spikes away with analog quantities. This close resemblance to biology makes our model well suited for guiding experimental research with high level computational issues. | |
Projects: | |
Publications:Distributed Synchrony (Zuohua Zhang and Dana H. Ballard)A Model of Predictive Coding based on Spike Timing (Dana H. Ballard and Rajesh P.N. Rao and Zuohua Zhang) A synchronous firing model of LGN (Dana H. Ballard and Zuohua Zhang) A Single-spike Model of Predictive Coding (Dana H. Ballard and Rajesh P.N. Rao and Zuohua Zhang) | |