No CPU required for integrating data from different senses

69 PD star-209371_1920Most of the time the brain receives information from different senses about the same object or event. For example, to localize an object that makes noise we use both visual and auditory information – if these are available to us -, process called multisensory information integration (MSII).

It is generally believed that the way this integration happens in a physical sense is by getting all these data in a special brain region dedicated to integrate information from multiple senses. And that there are several regions like that in the brain; for example superior colliculi integrate the visual and auditory information from the example above. This belief is not without empirical support. Indeed, many experiments both in vivo (i.e. in the awake behaving animal) and in silico (i.e. simulated on a computer program by building neural networks) have strengthened this idea.

But Zhang et al. (2016) claim there is some data that doesn’t fit the model. So let’s build a different model. Which is exactly what they have done using continuous attractor neural networks (CANN) as the building blocks for a neural network that seems to be biologically realistic. The output of their experiments shows that instead of having a central processing area in the brain that integrates data from multiple senses, optimal processing can be achieved by a decentralized network with neurons that are reciprocally connected.

Today, uncharacteristically, I am covering a paper that I have not read in its entirety. That is because, frankly, they lost me after the first Methods paragraph where they describe the way they built the neural network. In my defense, I don’t think there are many people in the world outside the neural networks field that can follow the mathematical formulae (see the Excerpt pic).

Excerpt from the Methods section of Zhang et al. (2016, doi: 10.1523/JNEUROSCI.0578-15.2016.)

Anyway, my gut instinct is that both hypotheses have merit in that the brain uses both specialized multisensory areas, like superior colliculi and decentralized, distributed reciprocal connections, like the model proposed by Zhang et al. (2016).

Reference: Zhang WH, Chen A, Rasch MJ, & Wu S (13 Jan 2016). Decentralized Multisensory Information Integration in Neural Systems. Journal of Neuroscience, 36(2):532-47. doi: 10.1523/JNEUROSCI.0578-15.2016. Article | FREE Fulltext PDF

By Neuronicus, 14 January 2016