Sunday, August 25, 2019

An embarrasing possibility

In reviewing how the cuttlefish might create a camouflage skin pattern to match a give environmental background, one possibility is that it is simply a display system that uses the intermediate neurons between the eyes and the skin to transfer input to output in the most direct way possible [whatever that is] or else to transfer the statistical properties of the input to the output.

Either way, the consciousness of the cuttlefish is not in question. I suppose it knows perfectly well what it is displaying on its skin. So its consciousness might be a sort of "occasionalism" that derives from it performing such a duplication process. Or perhaps it's display is more like a reflex that is not willful and has no auxiliary meaning.

All of which gives rise to the question of how my consciousness is different from a display system, as per a squid? Can I legitimately claim to be doing more? Well yes there is certainly a delay between input and output.

While I am moving anyway

I am motivated to do X but not enough to move and do it. But then I am also motivated to do Y and do begin to move. But, where efficient, I do X first.
Probably related to this.

Thursday, August 15, 2019

To Giles Laurent, re cuttlefish camouflage algorithms

We went to a great lecture in Woods Hole this summer by Giles Laurent about how cuttlefish can match a background texture with the colored dots (chromataphores) of their skin. It was particularly fascinating in its discussion of image texture statistics being matched rather than image shape statistics.

The speaker distinguished between statistics of "things" (shape) and of "stuff" (texture) and this got me thinking about how the Blaschke problem of reconstructing shape from statistics was never evaluated for reconstructing texture from statistics. After all, the cuttlefish seems to be performing such an inversion in displaying an image on its skin that has similar texture statistics to its background. But look at the previous post about arrays of colored dots being building blocks of a form of texture recognition. There is an algorithm there, which is:

  • find an array scaled correctly to best match the background - for each of several color channels.
  • localize a piece of that array and display it with a matching color channel chromatophore
  • do this for all color channels.
So Giles, here is an algorithm: match the scale of different colors, locally. And synchronize the displayed colors with the perceived ones - either one channel at a time or with a mixing matrix that varies locally.

Trimming the hedge and visual cognition

The hedge has little green tips of growth. I think of it, late at night on a day when I was hedge trimming, and it devolves into an array of light green dots - as though the 'thought' or 'image' of those green growing tips is composed - in part - by detecting such an array. A small field of view pattern is recognized as a sample from a full field array. It does not matter which part of the full array is sampled for the array to be a recognition factor.
Also briefly in this fleeting thought, different arrays at different scales appeared to my mind's eye. With all those constant full field options, as a basis, in various colors, you would be able to recognize quite a bit.
Update: what this suggests is that texture recognition can be done using linear regression, not with the parameters of a straight line, but with the parameters of the grid of colored dots.