I plan to graduate this summer (July, 2010). Therefore, I am actively searching for a job. My long term goal is to understand the unsupervised learning process in the brain. So far, my training defines me as a computational neuroscientist; I have studied neuroscience, computation theory, signal processing, machine learning, dynamical systems, information theory and spike train analysis. Working with collaborators, I have done experiments on (rat cortical) neural cultures, analyzed lobster olfactory population coding, and studied discriminability of noise stimulation in rat auditory system. To continue my career, I am looking for opportunities to work with systems neuroscientists. With my strong theoretical background, I would like to analyze data, develop robust methods, design experiments for in vivo experiments, and formulate neural coding and learning principles.
Brain is intrinsically noisy, yet robust. How information is coded is strongly limited by what kinds of noise is present in the system. In the neuron’s point of view, if two incoming signals cannot be discriminated well due to noise, they must be treated as being similar. If the brain is highly optimized to process signals, a good signal similarity measure and noise model would coincide. My short term goal is to extend and apply the statistical analysis methods for spike trains to multivariate in vivo data. This will be a stepping stone for understanding neural code, and deriving plausible unsupervised learning algorithms.
Another aspect of neural coding in the brain that is not widely studied currently is the context dependence. It has been observed that individual neurons are not tuned for just one task, but participate in multiple tasks in a non-stationary manner. It seems that the neural ensemble is formed depending on the dynamic state. Compared to the widely used static neural tuning analysis, a dynamic neural code analysis is a much more challenging and data demanding task due to the higher degree of freedom in modeling. Using dynamic modeling techniques from machine learning, and signal processing, I propose to analyze context dependent population neural code, and decode using Bayesian filter like techniques. The similarity that can be induced from the variability of neural signals can be used to induce a Hilbert space such that kernel based algorithms can be efficiently implemented for modeling.
I’m open to both faculty and postdoc opportunities in US and Europe. Please contact via email |memming|at|cnel|.|ufl|.|edu| for more information.
No comments:
Post a Comment