Wesley R. Elsberry
Posts: 4991
Joined: May 2002
|
From a 2011 review article on neural models of grid and place cells:
| Quote |
[...] We shall begin by describing strengths and limitations of the first generation of grid cell models—models for formation and transformation of grid signals that were proposed during the first 1–2 years after the discovery of grid cells in 2005. We will then show how limitations of these initial proposals, as well as new experimental data, have inspired the evolution of a second generation of models during the past 2–3 years. [...]
A number of computational models have proposed mechanisms for grid-like firing patterns. These models have constrained the number of potential biological mechanisms for the grid pattern, and they have allowed the systematic investigation of parameters required for formation and maintenance of periodic spatial firing during irregular behavior. In this section, we shall summarize and compare these models and show how they have evolved in response to theoretical and experimental analysis.
Models of grid cells should capture cardinal features of grid cells such as the generation of a periodic spatial signal, the persistence of such periodicity in the presence of changing running speed and running direction, the variability of spatial periodicity within the cell population, and the presence of patterns of temporal structure such as phase precession. Models that satisfy all or most of these criteria historically fall into one of two classes, although some convergence has taken place more recently. The first class, referred to as oscillatory-interference models, uses interference patterns generated by multiple membrane-potential oscillations to explain grid formation. The instantaneous frequencies of the oscillators are determined by the running speed and running direction of the animal such that a spatial rather than temporal firing pattern is generated (O’Keefe and Burgess, 2005). The second class of models, referred to as attractor-network models, uses activity in local networks with specific connectivity to generate the grid pattern (Fuhs and Touretzky, 2006; McNaughton et al., 2006). Here, patterns of activity are moved across a network of recurrently connected, periodically active neurons in proportion to the speed and direction of the animal’s movement. Thus, grid patterns emerge by path integration of speed and direction signals in both classes of models, but the mechanisms for obtaining triangular periodicity are different. Models of each class have now evolved beyond their first iterations, to address criticisms and integrate experimentally demonstrated features of the grid cell population.
|
There are two broad points to be made about that. First, it sets out what are essentially the minimal set of attributes needed in a grid and place cell model in order to be considered biologically plausible. Second, it recounts the general characteristics of two classes of existing models that accomplish implementations of those features. Can Gary demonstrates that his GridCellNetwork.frm code from PSC already has most or all of the attributes described? (Phase precession is one of those, and the GridCellNetwork.frm file appears to lack any mention of "phase" anywhere.) We will need lines numbers for each attribute's implementation. What excuse does Gary have for ignoring the existence of the classes of models the review article discusses for his claim that he "only model[s] what has never been modeled before"? Note that models are specified as existing by 2007.
Here's an interesting observation from that review:
| Quote |
The original model predicted a decrease in the frequency of single-cell membrane-potential oscillations along the dorsoventral axis of MEC, in parallel with the decrease in the spatial frequency of the grid (O’Keefe and Burgess, 2005). Such a frequency change was subsequently demonstrated in whole-cell patch-clamp recordings of medial entorhinal layer II neurons (Giocomo et al., 2007).
|
A consequence of a particular model was tested and found to actually reflect the attributes of the biology. Gary's GridCellNetwork.frm files lacks any mention of anatomical orientation vis "dors" or "ventr", and "spatial" nowhere seems to be paired with "freq".
Contrast the discussion of prediction and test from the review article with this disclaimer from Gary's GridCellNetwork.frm file, lines 287-288:
| Quote |
' How our brain or other cognitive system might produce and combine signals into such a spatial representation does not
' matter to this model for demonstrating what is possible using the cell types that are known to exist in animals.
|
The review article notes that the noiseless versions of a class of model were criticized for leaving out relevant features of the biology:
| Quote |
Recently, multiple criticisms of the first generation of oscillatory-interference models have been raised. For example, several papers have criticized the oscillatory-interference approach for modeling biological oscillators as perfect sinusoids (Giocomo and Hasselmo, 2008a; Welinder et al., 2008; Zilli et al., 2009). In contrast to the modeled oscillations, in vitro slice recordings indicate that membrane-potential oscillations show a high degree of noise (Dudman and Nolan, 2009; Zilli et al., 2009), variance in frequency (Giocomo and Hasselmo, 2008a), and significant attenuation in high-conductance conditions, which may occur during realistic in vivo levels of synaptic input (Fernandez and White, 2008). Computational simulations indicate that accumulating noise interferes with the grid pattern.
|
Gary's GridCellNetwork.frm lacks any mention of "nois".
The review article discusses another assumption:
| Quote |
In addition, criticism has focused on the assumption that multiple, separate oscillations combine in the soma while maintaining independence in the dendrites (Remme et al., 2009). Successful implementation of models that rely on this assumption depends heavily on the ability of independent oscillations in different dendrites to unidirectionally influence the global, baseline oscillation. Using an idealized and detailed biophysical model based on sine waves, Remme et al., (2010) demonstrated that a biologically realistic bidirectional interaction between the local dendritic oscillations and global oscillations (in this case, soma oscillations) results in complete phase locking between all oscillations and a subsequent loss of the grid cell firing pattern. Phase locking occurred in the range of hundreds of milliseconds, even with parameters generously skewed toward promoting dendritic independence (Remme et al., 2010). Though not ruling out the potential importance of oscillatory and resonant properties, the detrimental effects of phase locking emphasize the importance of multicellular and network mechanisms in the generation of spatial periodicity.
|
Gary's GridCellNetwork.frm fails to mention "phas" or "lock" anywhere within it.
Of course, Gary could counter that he is doing an attractor network model, and the specific criticisms above are of oscillator-interference models. But phase precession is raised as a topic for attractor network models, too:
| Quote |
One major limitation of the initial attractor models for grid cell formation was the lack of temporal dynamics that could contribute to phase precession in grid cells (Hafting et al., 2008).
|
Again, Gary's GridCellNetwork.frm file shows no sign of taking any note of phase precession.
The review notes a critical feature of attractor network models:
| Quote |
Finally, it is worth noting that the validity of the attractor models relies on the assumption of specific connectivity between grid cells with similar spatial phase.
|
Given that Gary's GridCellNetwork.frm file lacks any mention of "phas" whatsoever, it seems a given that Gary's "model" doesn't have this feature that the review says is fundamental to model validity. Of course, Gary could rebut that with line numbers where the implementation of multiple scales of spatial phase can be found in GridCellNetwork.frm. On past form, I expect Gary's reply will feature "creep", but not in any context related to organismal motility.
--------------
"You can't teach an old dogma new tricks." - Dorothy Parker
|
