Uri Rokni’s Home Page

Department of Brain & Cognitive Sciences, MIT
Howard Hughes Medical Institute



Brain & Cognitive Sciences Dept.

MIT, 46-5065

Cambridge, MA 02139

voice: 617-324-3746

fax: 617-452-2913




I am a postdoc working in the Seung lab. I received my BSc in physics and mathematics, MSc in physics and PhD in physics in the Hebrew University in Jerusalem. In my MSc thesis I studied how a 2D nonlinear oscillator is entrained by an external oscillatory input. In my doctoral studies I switched to the field of computational neuroscience. In particular, I studied how neural computations are performed by the intrinsic dynamics of recurrent neural circuits, in both sensory and motor systems.



I’m interested in how the brain improves its performance with experience. In my postdoc I chose to focus on motor learning. It is believed that motor learning occurs by converting sensory feedback on the system’s performance into appropriate synaptic changes in the motor system. However, the nature of this transformation is still largely unknown. A number of research groups have started to investigate this question experimentally by measuring how neural activities in the motor system change when an animal learns a motor task. I am working in collaboration with experimental labs and developing theories which help interpret the results of such experiments and suggest new experiments.

A theme which is emerging from my postdoctoral research is that the redundancy of the motor system has a profound impact on how the neural representation is shaped by learning. By redundancy, I mean that many different neural representations can give rise to the same behavior. Such redundancy results from the vast convergence from many neurons in the motor system to few motor outputs. In collaboration with Emilio Bizzi from our department, we found that redundancy can explain an intriguing finding in their data – the neural representation in motor cortex is continually changing, even when practicing a familiar task. A theory I have developed shows that such unstable neural representations emerge in redundant networks with plastic synapses (this work will be published soon in Neuron). In another collaboration with Michale Fee from our department, we found that redundancy can explain another intriguing phenomena in the song motor system of the zebra finch – the firing patterns in RA, an area involved in song generation, are time locked to the song but very weakly correlated with acoustic features. Using simple models of the song motor system, we found that neural representations which are weakly correlated with the motor outputs are formed, if RA is redundant and the learning process is noisy. In the near future, I plan to further explore the consequences of redundancy in neural circuits.



· U. Rokni, A. G. Richardson, E. Bizzi, H. S. Seung, Motor learning with unstable neural representations, Neuron, 54:653-666, (2007) (pdf) (Supplementary material - pdf)

· J. A. Goldberg, U. Rokni, T. Boraud, E. Vaadia, H. Bergman, Predicting spike correlations from local field potentials in the cortico-basal-ganglia circuitry of normal and parkinsonian primates, J. Neurosci., 24(26):6003-10, (2004). (pdf) (Supplementary material - pdf )

· J. A. Goldberg, U. Rokni, H. Sompolinsky, Patterns of Ongoing Activity and the Functional Architecture of the Primary Visual Cortex, Neuron, 42:489-500 (2004). (pdf)

· U. Rokni, E. Vaadia, O. Steinberg, H. Sompolinsky, Cortical Representation of Bimanual Movements, J. Neurosci., 23(37):11577-86 (2003). (pdf)

· D. D. Lee, U. Rokni, H. Sompolinsky, Algorithms for Independent Components Analysis and Higher Order Statistics, Proc. of the 1999 Conf. on Advances in Neural Information Processing Systems 12. (pdf)

· U. Rokni, L. Friedland, Double Autoresonance in Two Dimensional Dynamical Systems, Phys. Rev. E, Vol. 59, pp. 5242-5253 (1999). (pdf)

Ph.D. Thesis
(2004): The brain as a generative dynamical system (pdf)