Review from Referee 2:

The authors present results suggesting that a neuron equipped solely with STDP can learn to detect a repeating spatio-temporal pattern that is buried into otherwise non-structured spiking activity with constant firing rates. The pattern can appear at random places and thus, no time reference for pattern detection is needed. The mechanism is said to be an emergent phenomenon of the STDP learning rule. The study addresses an important problem of the detection of the spatio-temporal patterns in neuronal spiking activity.

It is my opinion that the study is incomplete because no sufficient evidence is provided to suggest that similar mechanisms take place in the brain. I have the following major concerns and comments:

- The neurons begin to learn by using too high firing rates: 64 Hz for the example neuron. And the performance drops with low firing rates. The value of 64 Hz is too high for most cortical neurons even when they are driven strongly. In the present system, this firing rate is supposed to be a property of non-specialized neurons that yet need to learn a new function and that are normally expected to fire at the level of spontaneous activity. These values are normally more than an order of magnitude smaller than 64 Hz. Therefore, it seems that the present mechanism is limited only to a small proportion of neurons that run with such high firing rates. This fact should be stated and its implications for the brain functions should be discussed.

- The authors cite no empirical evidence that the patterns similar to the simulated ones exist in the brain.

- What happens if multiple patterns are interleaved within the responses? If a neuron can learn a pattern reliably only when a single pattern is inserted into the activity that is otherwise always random, it is not clear how a practical use of such a neuron can be conceived. The system should be able to distinguish an arbitrary number of different patterns.

- The latter issue seems to be implied in the discussion (page 7, second paragraph) where it is suggested that self-organization based on varying thresholds can lead to robust learning. The arguments underlying these statements are not obvious and thus, the statements should be supported by proper theoretical analyses and simulations.

- Some parameters used in the example simulation seem to be biologically implausible or at least at the edge of the plausibility range. For example, it is not clear why the jitter of 1 ms would be sufficient to mimic the variability of neuronal responses.
- The report of the analysis of the parameter space (i.e., Figure 7) is necessary but is insufficient in the present form. These results should constitute a major part of the paper and here they have been compressed into only a single paragraph. Much more discussion is needed! In particular, the simulation results should be related to the biological values of the parameters. We need to know the conditions under which the proposed mechanism can operate in the brain. Also, it would be nice to identify the parameter that matches the biological values LEAST and to discuss why the brain should nevertheless evolve to use this mechanism.

- Additional parameters could be investigated. For example, what happens if the membrane time constants are different than 10 ms? What happens if the patterns are shorter/longer?

- On the positive side, I think that the authors can be more relaxed about the convergence time. Even if the neurons learn the patterns within much longer time periods than the 2000 discharges or so that were needed in the reported simulations, the results can be taken as biologically plausible (e.g., perceptual learning can take long time). It would be great if long learning times could enable even the neurons with low firing rates to learn the patterns.

- I would suggest to the authors to try to answer the following question: Given the known properties of neuronal activity in the cortex and given the known properties of the pyramidal cells, can pyramidal cells use the present mechanisms to learn and detect the patterns embedded in the cortical activity?

----------
N.B. These are the general comments made by the reviewer when reviewing this paper in light of which the manuscript was revised. Specific points addressed during revision of the paper are not shown.