Computational Model Linking Neural Activity to Behavior

Question

A big question in neuroscience is how neural activity represents knowledge. We can use modelling to explore how different levels of neural activity- subthreshold currents, action potentials, local field potentials, etc.- relate to one another and, finally, to behavior.

Some models simulate behavior, for example which color flowers a bee prefers or how a mouse will navigate the Morris water maze. (Think Ch. 9 in Dayan & Abbot (2001).)

Other models simulate the behavior of groups of neurons. Some of those even address what knowledge that activity could be representing. (Think Ch 10 in Dayan & Abbot (2001).)

What are examples of linking the two? (I'm aware of computational limitations, aside from computing power such as the foolishness of simulating stiff equations for long periods of time. )

EDIT: One example is exploring how local field potentials (LFPs) could arise from networks of integrate-and-fire neurons. Earlier models of LFPs were phenomonological. They were simply coupled oscillators with different time constants and so didn't allow one to see how more detailed levels of description could create LFPs.

Answer 1

There are several such models in the field of auditory perception. For example Patterson 1996 ^[1] suggests a model that starts with a simulation of the cochlea and the neural activity and reaches up to perception; Winkler 2006 ^[2] reviews the process of auditory perception, again from the cochlea up to perception.

Somewhat old and does not mention a specific model, but still a very relevant review of the neural activity - behaviour interaction debate is provided in Barlow 1972 ^[3].

---

[1] Patterson R.D., Holdsworth J., A functional model of neural activity patterns and auditory images‏. Advances in Speech, Hearing and Language Processing, 1996. link.

[2] Winkler I., Denham S.L., Nelken I‏., Modeling the auditory scene: predictive regularity representations and perceptual objects. Trends in cognitive sciences, 2009. doi: 10.1016/j.tics.2009.09.003

[3] Barlow B.H., Single units and sensation: A neuron doctrine for perceptual psychology? Perception, 1972. link

Answer 2

The only modelling method that I know of for creating large-scale biologically based models is the Neural Engineering Framework (NEF).

The NEF is basically a framework for associating functional computations and dynamic systems to biologically plausible populations of neurons. Given this foundation, advanced applications linking behaviour to neural function become much easier.

An example of this is Dan Rassmussen's work on hierarchical reinforcement learning, which relates behaviourist reinforcement learning to neural activities and brain regions as observed in neural data.

Spaun, which is an amalgamation of many isolated models (basal ganglia, visual cortex) created using the NEF, has also replicated various human behaviour (same rate of forgetting, recognition and recall) in simple tasks such as list memorization.