On this page you will find a few examples from my recent research.

Neocortex simulating a microcircuit of 10000 spiking neurons reacting to visual gratings. This is a study trying to assess the possible mechanisms of brightness induction. On the left side you can distinguish the neural maps and firing of various neurons (white dots). The activity in the microcircuit is self-sustained (not dependent on the input) and there is obvious variability in the peri-stimulus-time-histograms (PSTHs) on the right side. Since there is no added noise to the system, the high variability (across a population of 25 neurons) suggests very complex dynamics of the network and noise as "network effect".

Precision in the brain matters! An artifficial spiking neural network is being stimulated by real neurons in cat area 17. A classifier is being trained to extract information specific to different conditions of stimulation from the artificial network’s activity. When the spikes in the recorded data are jittered, even by very small amounts, the classification performance drops drastically, suggesting that the extracted information is encoded not in the rates (not affected by jittering) but in the precise timings of the cortical neurons.

Simulation of a large scale microcircuit with Neocortex. The model is highly recurrent and self-sustained, displaying complex non-trivial dynamics.

Description of the dynamical state space of a self-sustained, isolated microcircuit (left). The system fills in the entire available state space, while its trajectory shows no obvious attractors of limit-cycles. When the microcircuit is stimulated with an oscillatory pattern, it “jumps“ on a limit cycle with variable periodic orbits, dependent on the oscillation’s properties (middle). A column from cat visual cortex shows highly compact state spaces (right).

Transitions induced into a small population of readout neurons by a freely evolving (left), oscillation stimulated (middle), brain recorded (right) microcircuit. Oscillations constrain the state space of the readout population, making it more predictable.

Dynamics of different neuronal models in response to current stimulation. Simulation carried out by Neocortex.

Synaptic dynamics of a NMDA type excitatory synapse. Simulation carried out by Neocortex.

Complex phenomenological model of spike-timing-dependent plasticity (STDP). Available in Neocortex as a model of synaptic self-organization.

Scalability of memory requirements in Neocortex for large networks. Results presented for two synaptic representation strategies: implicit and explicit representations.

A critical analysis on the relevance of rate responses and average rate profiles. The simulation reveals inconsistent extraction of information as one tries to analyze cells that are high in a processing hierarchy. As recorded cells get further away from the sensorial areas, the relevance of rate responses is decreased, suggesting the need for different strategies of decoding neural information.