In complex system view, the last question means that different abilities of brain are "emergent property" of neuronal network as a complex system. You can search for this very useful topic (it's related to memory, synchronization, epilepsy, ...).

I see. However the point is: to understand the emergent property of a neuronal network, do we really need data with very fine temporal resolution? It is like asking whether we need to know the behavior of each single atom of a gas to understand pressure, temperature etc. If the answer is 'yes', then 'slow' techniques - from calcium imaging up to fMRI- cannot give much insight into brain machinery, and I am quite uncomfortable with this conclusion.

There is no compelling reason to believe that there is one 'real language' of the brain. Here is a quick listing of instances from sensory neuroscience, especially visual neuroscience. Considerable spike-timing precision has been observed in sensory pathways, notably the early visual pathway [1] and more recently, accounting for fine-grained spike-trains has been shown to be required to account for behavioral performance in mammals [2]. Information-theoretic approaches such as those used in this study[3] show how decoding models which assume population coding lead to reliable estimates comparable to those attained behaviorally. Here is a useful review paper highlighting the diversity of neural coding [4], precluding any single catch-all description at least in currently prevalent frameworks.

[1] https://www.researchgate.net/publication/11210450_Precise_firing_events_are_conserved_across_neurons

[2] https://www.researchgate.net/publication/24211554_Ruling_out_and_ruling_in_neural_codes

[3] https://www.researchgate.net/publication/49739515_Decoding_the_activity_of_neuronal_populations_in_macaque_primary_visual_cortex

[4] https://www.researchgate.net/publication/40833298_Sensory_neural_codes_using_multiplexed_temporal_scales

Article Precise Firing Events Are Conserved across Neurons

Article Ruling out and ruling in neural codes

Article Decoding the activity of neuronal populations in macaque pri...

Article Sensory neural codes using multiplexed temporal scales

In a couple of earlier answers the essential principle has been highlighted: nervous system processes should unfold and allow transfer of conrtrol signals at a time scale compatible with how fast meaningful (and potentially dangerous or highly useful) changes in environment happen. This is both from the afferent, sensory perspective and motor control related perspective. Fast moving objects (falling rocks, tennis balls, attacking snakes) change location/direction in the range necessitating millisecond diapasone resolution for sigbnals in the NS. Furthermore, necessary (neuro)biological processes unfolding at a slower time-scale can be useful for adaptation insofar as they can be meaningfully and substantially related to the faster processes in the hierarchical mechinery that needs mechanisms of timing. This means that there should be some compatibility and inter-level communication between the timing systems. Slow processes have to be related to fast processes and the time scales should be commensurable. In most of the efficient systems the dynamics and adaptive value are advanced and superb, provided that the staple, typical range of the speeds with which periodic processes unfold is not the upper limit of the temporal resolution of the system, but there is a considerable reserve of temporal resolution at higher scales (compared to the typical range). So if environmental changes an organism has to cope with unfold at the timescale of tens of milliseconds, some lower level NS systems must have resolution perhaps even at the sub-millisecond level. Thus we have a voluminous, hierarchical, multilevel and coordinated system of timing the transfer and emergence of signals spanning from sub-ms up to seconds, hours, months, and beyond.

Thank you all for the aswers.

To summarize: it is worth recording single spikes (millisecond time-scale) because these are related to stimulus-response observations in sensory systems.

I suspect that this 'fast' neural code is not useful to "transfer conrtrol signals at a time scale compatible with how fast meaningful changes in environment": the transfer of information is slow at the level of synapses (about 0.5 millisec), so that after few hundreds neurons the 'millisecond' neuronal code would be completely changed.

In fact, even in sensory systems, reference 4 posted by Arwin suggests the co-occurrence of other codes at slower time-scales (multiplexing).

If you have time, please see the video:

http://www.youtube.com/watch?feature=player_embedded&v=Vo3TchUQM1w

and the original reference:

http://www.cell.com/current-biology/retrieve/pii/S096098221300002X

There we see some kind of neuronal code, but the timescale is slow: this is calcium imaging, not electrophysiology.

FInally, I see everyone agrees that fMRI is not giving informations about any neuronal code.

Logothetis spent considerable time and energy contrasting and explaining differences underlying electrophysiology and fMRI. Explaining LFPs, MUA, single units. This was the beginning of their undertaking recording electrophysiology, while fMRI scanning... A very technically challenging problem.

Following their results and of others arguing for more comparable tasks between electrophys in non-human primates and fMRI in humans , we in the Suzuki Lab and Craig Stark's lab similarly analyzed two data sets in an identical fashion across a near identical task performed by humans undergoing Bold fMRI and rhesus monkeys recording from hippocampal and Entorhinal areas.

Hargreaves et al., 2012, Neuron.

If we take for instance a simple example about what happens in neural response at the level of ganglion cells: the spatio-temporal filtering --> with a constant non-moving stimulus, the ganglion cells do not answer anymore after of couple of tens of milliseconde. Another cell recording techniques wouldn't be suitable to explore the speed of variation in response of the neural networks.

In this paper:

http://www.nature.com/nature/journal/v440/n7087/full/nature04701.html,

you see that the author did a binning (sampling period) for the spike at 10 ms. For sure he won't go far from that because of he beleives that if you go beyond 10-20 ms ... you can't anymore make such analysis.

I am currently developing tools to perform statistics on spike train (https://www.researchgate.net/post/What_kind_of_computational_software_tools_does_neuroscience_community_need) in order to make advances toward the study of neural code and actually, the basic sampling period that we use is 1 ms, but we can configure analysis with 10-20 or another sampling period.

Thank you Milan ... Looking forward to hear from you.