7

In a recent review article (Arnal & Giraud, 2012), the authors delve into changes in cortical oscillations which assist in predicting the causes of a sensory stimulus (the "what", via predictive coding) and its timing (the "when", via predictive timing).

This allows the cortex to dampen its response to a repeated and/or predictable stimuli so that more "processing resources" can be reserved for other less predictable (and potentially more important) stimuli.

Predictive coding can be modeled using a Bayesian framework, with predictions being made about the "what" aspect conditioned on the reactions to prior stimuli.

In certain circumstances --- intense military march to a drum beat or, for the young, dancing to a bass-heavy song in a nightclub --- it seems to be possible to dampen the "what" aspect completely since the Bayesian priors on "what" are near-certainty. To what degree can the brain move resources from the "what" to the "when" to achieve a precise level of timing for conversion between sensory and motor output?

Arnal, L.H., Giraud, A-L. (2012) Cortical oscillations and sensory prediction. Trends in Cognitive Sciences 16(7), 390-398.

Artem Kaznatcheev
  • 12,323
  • 5
  • 70
  • 164
Chuck Sherrington
  • 7,112
  • 4
  • 43
  • 83
  • Presumably when the 'what' is entirely predictable, but the 'when' is not? Like the oh so annoying sound of a dripping faucet :) Would that be an example? I'm not entirely clear what the question is though, but I plan to read this paper in the near future and then it might become clear. I don't think you can ever shut down the 'what' completely though. There's always an input layer (i.e. prediction error layer) which gets excited, and subsequently dampened before progressing to the next stage of processing. But it does get excited on input. – Ana Jul 01 '12 at 20:12
  • Hm, I now realize you're assuming more resources are put into processing something predictable, so my leaky faucet is the exact opposite of what you're asking. In general, there is more neural activity to unpredictable events, i.e. more resources are used. I'll get back to your question once I read Arnal's new paper! – Ana Jul 01 '12 at 20:17
  • @Ana A great point. I think the annoying sound would be an example of something in which the "what" is still salient (for whatever reason, as the stimulus is difficult to tune out). I'm more going for an example, say, when someone is jogging to a song with a prominent bassline, does the actual nature of the music stimulus get dampened to a point where all that matters is the timing of the pulses, with the fact of it coming from a 'drum' gets factored out completely? – Chuck Sherrington Jul 01 '12 at 20:21
  • 1
    So this question is not about dubstep? – Artem Kaznatcheev Jul 09 '12 at 04:08
  • @Ana I was wondering if, in the meantime, you had any other resources on this subject that might be helpful? – Chuck Sherrington Aug 09 '12 at 04:11
  • @ChuckSherrington: I just got back from holidays and put the Arnal paper in my bag to read while commuting, with this topic in mind :) I'll get back to it soon. In the meantime, the book 'On Intelligence' by Jeff Hawkins is a very good introduction into the brain as a predictive machine. I'm not sure how deep you want to go, and in what direction exactly. I don't know much about timing, and reading myself into this literature is the plan for the next half year. I hear that Kia Nobre has many papers on the subject. – Ana Aug 09 '12 at 09:31
  • @ChuckSherrington - Btw on my profile page here there's a link to my webpage, from which you can find the main page of the lab group I belong to. We put published papers online, so you can find some stuff to download, but we're a fairly new group, so not so many publications yet. The group is called 'Prediction and attention' - there might be a few things directly interesting to you. – Ana Aug 09 '12 at 09:35
  • @Ana Wow, thank you! I'll check out all 3. No need to rush, I know you are likely very busy. :) – Chuck Sherrington Aug 09 '12 at 09:48

1 Answers1

4

To what degree can the brain move resources from the "what" to the "when" to achieve a precise level of timing for conversion between sensory and motor output?

I think that you may be making a false assumption here. I don't think that the 'what' and the 'when' are in competition for resources. According to Arnal and Giraud (and many others), having a valid prediction of when an event is going to occur and having a valid prediction on which event is going to occur, both lead to reduced primary sensory activity (and to a shortened reaction time). As such, you could say that the resources that go into making both types of predictions (the 'when' and the 'what'), reduce total resources used. In the opposite situation, where there is an invalid prediction, there is not only more total neural activity in primary sensory areas, but stimulus representations are also less sharp (Kok et al., 2012).

Furthermore, the differences seen in frequency space that are related to 'what' and 'when' predictions, as well as the fact that 'when' predictions seem to have a necessary stimulus-driven component while 'what' predictions do not, also point to a possible dissociation in resources used. In other words, it might not be the same neurons firing anyway. As an example, Wacongne et al (2012) suggested a model on how the feedforward and the feedback portion of predictive coding (of 'what' predictions) might work using different neurons within a single cortical column.

Finally, less resources does not necessarily translate to better sensory-motor conversion: selectively attending to a stimulus increases the amount of sensory brain activity, but decreases reaction times. It seems that it might boil down to how sharp the stimulus representation is (which is connected to valid/invalid 'what' predictions), followed by how strongly these neurons are firing (which is connected to selective attention). Squandering resources would mean having the wrong neurons firing, rather than firing more vs. less in a given sensory area. Once again, this would mean that forming any type of prediction leads to better use of total resources.

I'm still not sure if this answers your question, as the examples you give (e.g. drum beat) contain certainty on both the 'what' (sound of drum) and the 'when' (time of sound onset). If you want to dissociate the two, you could have a situation where you know what is going to come (sound of drum) but with a jittered stimulus onset asynchrony. Or, you could have a precise timing with novel sounds (a sound of a different pitch is displayed every half a second). In those cases your predictions would be mainly on the 'what' vs 'when', but in both cases hearing any of these sounds would carry an element of surprise, and would therefore lead to more primary sensory activity and a longer RT, i.e. more resources would be used.

References:

Wacongne C, Changeux JP, Dehaene S (2012) A Neuronal Model of Predictive Coding Accounting for the Mismatch Negativity. The Journal of Neuroscience 32:3665-3678.

Kok P, Jehee JFM, de Lange FP (2012) Less is more: Expectation sharpens representations in the primary visual cortex. Neuron 75:265-270.

Community
  • 1
Ana
  • 2,553
  • 16
  • 20