Research

My interest is to understand how does the brain process the stream of information it constantly acquires. To this end, I'm looking at groups of neurons in visual cortex and trying to make sense of how they represent visual information, particularly via spatio-temporal spike patterns. My work follows two main directions. One deals with the development of tools for analyzing and visualizing high dimensional neuronal data. The other identifies which properties of the neuronal activity carry most information and how does the representation change over time, as a function of learning.


Visualizing Multineuronal Activity Patterns

Spike rastergrams are an efficient tool for visualizing the activity of simultaneously recorded neurons. However, the visual detection of spike patterns across multiple neurons is largely dependent on the arrangement of neurons in the rastergram. The Gestalt principles that govern our visual system dictate that spike patterns distributed across neurons that are not neighbors in the rastergram will most likely be missed.

To enable the visual identification of multineuronal spike patterns, we devised a method to color them based on their reciprocal similarity. For example, if pattern A is similar to pattern B but different from pattern C, then pattern A and B will be assigned similar colors (e.g., red and orange) while pattern C will be assigned a different color (e.g., blue). This way, color becomes a signature of pattern identity and the occurrences of similar patterns can be easily visualized across entire recording sessions.


Figure. Transforming simultaneously recorded spike trains into color sequences. For each time window, the spiking activity of all neurons is mapped onto a color space using Kohonen self-organizing maps.

The labeling of spike patterns with colors enables the visualization of each trial as a sequence of colors. Grouping several color sequences by a certain criterion (e.g., stimulus, time of recording) can reveal regions in which similar patterns (colors) occur consecutively along the trial and/or consistently around the same time points across different trials.

Figure. Color sequences corresponding to trials recorded with the same stimulus
(a drifting sinusoidal grating). The three yellow stripes indicate the presence of similar activity patterns at specific time points across all trials.

This method enables the intuitive visualization of neuronal population dynamics and enables the identification of periods of interest, which can be further subjected to more quantitative analyses. Although it was designed for the visualization of multielectrode spike trains, the method could be applied also to simultaneously recorded continuous signals (e.g., LFP, EEG, MEG).

Article links:  [abstract]  [pdf]  [source code]


Timescale of Informative Multineuronal Activity Patterns

A fundamental and much debated problem in neuroscience is how neurons in the visual system work together to encode information that is sampled by our eyes.
To address this problem, we recorded spiking signals from the primary visual cortex of anesthetized cats while the animals were presented with stimuli of various temporal dynamics. These data were then fed to a set of classifiers that were able to identify and quantify the occurrence of multineuronal spike patterns on arbitrary timescales. By considering different properties of spike pattern occurrence (e.g. stimulus specificity or stimulus time-locking), classifiers could determine the timescales that were most informative for discriminating between visual stimuli.

Figure. Discriminating a visual stimulus from a set based on the activity of 22 neurons at various timescales. Stimuli were movies showing natural scenes. The colored arrows indicate moments in time where one timescale was more informative than the others in discriminating the stimulus identity.

Our results indicate that the internal timescale of the brain, i.e., the time window required by neurons to encode a given aspect of the visual stimulus, is tightly correlated to the external timescale of the visual stimulus, i.e., the speed with which visual images on the retina change. Thus, when quick responses are needed as a reaction to the attack of a predator, rapid changes in the field of vision can trigger fast neuronal responses that convey information rapidly, on timescales of a few milliseconds. This suggests that the brain is well adapted to the environment, matching the speed of its internal activity to the constrains imposed by the environment.

Article links:  [abstract]  [pdf]