Nimrod Shaham

Hebrew University of Jerusalem | HUJI · Racah Institute of Physics

Humans have the remarkable ability to continually store new memories, while maintaining old memories for a lifetime. How the brain avoids catastrophic forgetting of memories due to interference between encoded memories is an open problem in computational neuroscience. Here we present a model for continual learning in a recurrent neural network comb...

During fixation and between saccades, our eyes undergo diffusive random motion called fixational drift. The role of fixational drift in visual coding and inference has been debated in the past few decades, but the mechanisms that underlie this motion remained unknown. In particular, it has been unclear whether fixational drift arises from periphera...

Humans have the remarkable ability to continually store new memories, while maintaining old memories for a lifetime. How the brain avoids catastrophic forgetting of memories due to interference between encoded memories is an open problem in computational neuroscience. Here we present a model for continual learning in a recurrent neural network comb...

During fixation and between saccades, our eyes undergo diffusive random motion called fixational drift. The role of fixational drift in visual coding and inference has been debated in the past few decades, but the mechanisms that underlie this motion remained unknown. In particular, it has been unclear whether fixational drift arises from periphera...

It has been proposed that neural noise in the cortex arises from chaotic dynamics in the balanced state: in this model of cortical dynamics, the excitatory and inhibitory inputs to each neuron approximately cancel, and activity is driven by fluctuations of the synaptic inputs around their mean. It remains unclear whether neural networks in the bala...

Network with random and sparse inhibitory connections between the sub-networks. A Projections of the nullclines m˙1=0 (blue) and m˙3=0 (red) on the m1 − m3 plane, based on Eqs 13 and 12. Here K = 1000, J˜=1.8. Insets show a schematic illustration of the nullclines near the fixed points, in which the angle between the lines is amplified for clarity....

Effects of correlated input noise. Results from simulations, for a network in which dynamical noise is added to the input E0. The noise has a correlation time of 3τ, which ensures that the correlation across neurons is not averaged out due to the asynchronous updating. The noise is described by an OU process: τnoiseξ˙=-ξ+σnoiseη(t), where τnoise =...

Mean squared displacement of location along the approximate attractor. (Same dataset as in Fig 4C in the main text.) A-B The mean squared displacement (MSD) of the location along the line for initial location X(0) = 0.05 (A) and X(0) = 0.08 (B) as a function of time. Error bars represent the standard deviation of the mean (black). Red: fit to an OU...

Non-mirrored connectivity. Results for a network in which the internal connectivity in each sub-network is drawn independently. A Population averaged activity projected onto the m1 − m3 plane. B Mean activities of the four populations: blue for one sub-network and red for the other. The higher activities are those of the excitatory populations (m1...

Working memory, the ability to maintain and use information for several seconds, is central to many functions of the brain. In the context continuous variables, an important theoretical model of working memory is based on neural networks, whose dynamics possess a continuum of marginally stable steady states. It has been unclear whether this theoret...

The simple interpretation of Pierre Auger Observatory ultrahigh energy cosmic rays (UHECRs) penetration depth measurements suggests a transition at the energy range 1.1-35×10^{18} eV from protons to heavier nuclei. A detailed comparison of this data with air shower simulations reveals strong restrictions on the amount of light nuclei (protons and H...

The simple interpretation of PAO's UHECRs' penetration depth measurements suggests a transition at the energy range $1.1 - 35 \cdot 10^{18} $ eV from protons to heavier nuclei. A detailed comparison of this data with air shower simulations reveals strong restrictions on the amount of light nuclei (protons and He) in the observed flux. We find a rob...