The isocortex of the mammalian brain has a remarkably regular structure [3,4,5]. Observational data suggests that the cortex is composed of groups of hundreds or thousands of neurons that form so-called mini- or macro-columns, respectivey [6,7]. This regularity is commonly seen as an indication that there could be a common computational principle underlying all cortical function. However, despite numerous suggested models (e.g., [8,9,10,11]) no such principle has been discovered to this date and some researchers question if such a principle exists at all [12,13].
I think that a number of initial hypotheses for the basic characteristics of such a general principle of cortical function can be derived from my previous work in which I investigated the computational properties of entorhinal grid cells.
In my grid cell research I was able to demonstrate that the activity of grid cells can be modeled by a general computational principle according to which a local group of neurons learn a sparse, distributed representation of an arbitrary input space that is shared by the group. This distributed representation is learned in an unsupervised fashion based on principles of Hebbian learning and local inhibition, and has a capacity that is exponential in the number of neurons.
My current research tries to apply what I have learned from modeling grid cells to the challenge of modeling cortical function. My core assumption is that the behavior of individual grid cell groups as described by my grid cell model is in fact a general behavior of neuronal groups throughout the cortex. Therefore, I approach the task of modeling cortical function by using local neuronal groups that learn a representation of their input space as the basic building block of my cortical model. First ideas of that work are published in two early papers [1,2]. In [1] I show that two neuronal groups can form an autoassociative memory cell that serves as a core element of a simple cortical column. In [2] I discuss notions of memory and computation in theories of cognition that are build on assumptions about the cortex that likely do not hold. My research in this field is ongoing and new results will be published here in the future.
References
1
,
A Grid Cell Inspired Model of Cortical Column Function,
In: 10th International Joint Conference on Computational Intelligence (IJCCI 2018), Seville, Spain, September 18-20., pp. 204–210, 2018,
[pdf|doi|bibtex]
2
,
Challenging the Intuition About Memory and Computation in Theories of Cognition,
In: Proceedings of the 11th International Joint Conference on Computational Intelligence - Volume 1: NCTA, (IJCCI 2019). INSTICC. SciTePress, pp. 522–527, 2019,
[pdf|doi|bibtex]
3
,
Cortical connectivity and sensory coding,
In: Nature, 503:51, 2013,
[doi]
4
,
Cognitive consilience: Primate non-primary neuroanatomical circuits underlying cognition,
In: Frontiers in Neuroanatomy, 5:65, 2011,
[doi]
5
,
Fundamental Neuroscience,
In: Elsevier Science, 2008,
[details]
6
,
The columnar organization of the neocortex,
In: Brain, 120(4):701–722, 1997,
[doi]
7
,
An organizing principle for cerebral function: The unit model and the distributed system,
In: The Mindful Brain, pages 7–50. MIT Press, Cambridge, MA, 1978,
[details]
8
,
Cortical columns: Building blocks for intelligent systems,
In: IEEE Symposium on Computational Intelligence for Multimedia Signal and Vision Processing, pages 21–28, 2009,
[doi]
9
,
Why does the neocortex have columns, a theory of learning the structure of the world,
In: bioRxiv, 2017,
[doi]
10
,
Rapid Processing and Unsupervised Learning in a Model of the Cortical Macrocolumn,
In: Neural Computation, 16(3):501–533, 2004,
[pdf]
11
,
A cortical sparse distributed coding model linking mini- and macrocolumn-scale functionality,
In: ArXiv e-prints, 2017,
[details]
12
,
The minicolumn hypothesis in neuroscience,
In: Brain, 125(5):935–951, 2002,
[doi]
13
,
The cortical column: a structure without a function,
In: Philosophical Transactions of the Royal Society of London B: Biological Sciences, 360(1456):837–862, 2005,
[doi]