Reader Comments
Post a new comment on this article
Post Your Discussion Comment
Please follow our guidelines for comments and review our competing interests policy. Comments that do not conform to our guidelines will be promptly removed and the user account disabled. The following must be avoided:
- Remarks that could be interpreted as allegations of misconduct
- Unsupported assertions or statements
- Inflammatory or insulting language
Thank You!
Thank you for taking the time to flag this posting; we review flagged postings on a regular basis.
closeSpatial or spatiotemporal patterns?
Posted by Dean on 10 Jan 2008 at 00:16 GMT
This is an interesting paper that addresses a fundamental issue in regards to both how the brain detects temporal patterns and the computational role of STDP.
However, it is stated that “… STDP is able to solve a very difficult computational problem: to localize a repeating spatio-temporal spike pattern embedded … ” and it is implied the the postsynaptic neuron is sensitive to temporal structure.
It is important to distinguish between spatial and spatiotemporal patterns, and I’m wondering if the postsynaptic neuron is actually only relying on spatial patterns and insensitive to temporal structure. The latency of the postsynaptic detector can be as short as 4 ms, thus, obviously, if there is a temporal pattern it is a very short one, less the 4 ms – probably 2 or 3 ms due to synaptic delay and the membrane time constant. Whether a 2 or 3 ms “pattern” constitutes a temporal pattern is up for debate, but it could be relevant to decoding “latency codes” (e.g., Gawne et al, 1996; Buonomano & Merzenich, 1999). Either way, a common test to determine if a neuron is detecting a spatio-temporal pattern is to reverse the stimulus. I suspect that if the authors reverse the 50 ms pattern the neuron would still fire, at the very end of the stimulus. Which by most criteria would prove that it is not sensitive to temporal patterns.
RE: Spatial or spatiotemporal patterns?
tmasquelier replied to Dean on 10 Jan 2008 at 17:08 GMT
As said in the discussion, after learning the neuron is selective to the nearly simultaneous arrival of spikes in the synapses STDP has potentiated (coincidence detection). It is true that shuffling the order of these spikes has no impact on the response. In particular, as you suggest, reversing the whole pattern would still make the neuron fire, this time at the very end of the stimulus, because the nearly simultaneous arrival of spikes that correspond to potentiated synapses would be preserved.
To be sensitive to the exact spike order, and not their mere simultaneity, additional mechanisms would be needed. For example, Feed-forward shunting inhibition can be used to make the postsynaptic neurone sensitive to order (see Thorpe S, Delorme A and Van Rullen R (2001). Spike-based strategies for rapid processing. Neural Networks). This kind of mechanism was not implemented here because we wanted to keep the situation as simple as possible, but it would certainly be interesting to add such features.
However, simultaneity is already a temporal aspect of the input, and the neuron is selective to it. For example. the pattern is hardly detected if a 4 ms jitter is added to it. We thus used the term "temporal pattern", while explaining the limit of our approach in the discussion: "there is no consensus on the definition of a spike pattern, and we admit ours is quite simple: here a pattern is seen as a succession of coincidences". This succession is another temporal aspect, which, even though not relevant after convergence, greatly facilitates the learning phase. It would be much harder to find a 4 ms repeating pattern on its own. But because the 4 ms is the start of a much longer sequence (here lasting 50 ms), the neuron only needs to fire somewhere within a 50 ms window to start learning the pattern, and tracking back through it until the first 4 ms are found.
Our claim is that if there is a repeating spatio temporal pattern in a spike train, STDP will find it and will track back through it until it finds the beginning of it, and NOT that after learning the neuron is selective to the whole spatio temporal structure of the pattern.