Rama Natarajan - Research

Research     Courses     Teaching

Reports     Misc     Home

Current Research

Neuronal Population Coding

The nervous system constructs internal representations of sensory stimuli according to constraints imposed by the architecture of the system and its functional abilities. Various aspects of input stimuli (for example, uncertainty in what is perceived) are encoded by a population of neurons in their selective response to the stimuli. There is substantial psychophysical evidence suggesting that the brain actually represents features of the input stimuli as probability distributions. In collaboration with my advisor Richard Zemel, I am studying how a population of neurons can optimally represent input signals with time-varying distributions.

Past Graduate Research

Neural Data Analysis at the Stomatogastric Ganglion Lab, Rutgers University, Newark, NJ, USA.
The lab studies the generation of rhythmic motor patterns responsible for chewing and digestion of food in the crustacean stomatogastric nervous system (STNS). The response patterns are dependent on mechanisms that operate on different time scales. The significance of the timing of these responses has been recognized as a means of neuronal information coding. To understand the fundamental mechanisms by which the rhythmic circuits generate and regulate patterns, the temporal dynamics of the neurons, as evinced in their burst activity, must be characterized.

With guidance from my advisor Dr. Farzan Nadim, I developed a computational tool for electrophysiologists to analyze the extracellular single-electrode recordings of multi-unit neuronal activity from pyloric network of crab Cancer borealis. The objective was to automatically detect the exact times of beginning and ending of patterned activity and analyze the phase differences in this response with respect to the activity of a pace-maker neuron. The software is an implementation of a model for unsupervised learning in which representations of distinct neuronal burst patterns, characteristic to different neuronal types, are learnt. This is then used to sort the recorded multi-unit spike train into responses from individual neuronal types, and then perform phase analysis.
Poster | Presentation