3 ms·

I think discrete pipelines is the wrong paradigm. I surmise what is happening is the incoming information is being changed to the frequency domain for much the

• fluoridation 4y ago
  Unlike hearing, I don't think the visual system works in the frequency domain. Common psychedelic visual hallucinations are better explained as either transformations in the space domain, or stages of a pipeline being reconnected in odd ways. As an example of the former, a sharpen filter can be implemented as a simple convolution, which would be compatible with crosstalk in the neurons processing the "pixels" of the visual input. Of the latter, fractals are a common theme in hallucinations; they can be explained as feedback loops in a pipeline, where the signal ends up making several loops through a few processing stages. The sharpen hallucination might also be caused like this. Perhaps the brain already applies a sharpen filter and what happens is that it's applied multiple times to the signal.
  • NoToP 4y ago
    I think you might have missed out on learning about the 2d Fourier transform and the meaning of "frequency domain". The standing wave patterns of the feedback loops are exactly the wave patterns you want to decompose the image data into. The waves have a characteristic frequency inversely proportional to the length of the loop. One can apply a decomposition into a wave basis just as easily on the space domain as one does on the time domain. The convolution in the original domain just becomes multiplication in the frequency domain. Sharpen and blur filters are literally just high pass and low pass filters applied to the 2d Fourier transform. I highly doubt there is one big pipeline to which the original sensory inputs go through some fixed filters and then return back to the main start point as feedback. It's not exactly like pointing your camera at the projector, or holding the microphone too close to the speaker. Rather, there's many possible paths for neural inputs to loop back, and none of these loops is the dominant "pipeline". It's more like a microphone is near a whole wall of speakers, each with its own characteristic delay and amplification factor. The repeating spatial patterns of whatever length directly translate to a standing wave in neural loops with the right amount of loop length. The transform to a convenient wave basis doesn't need to be implemented as a layer in a pipeline because the loops themselves are the transform. A transform into frequencies (in some weird basis) is almost an unavoidable consequence of connecting neurons at random. Just like any old metal sheet is a bell if you hit it with a hammer, any old clump of neurons is going to have standing waves of activations that resonate with some sensory input pattern. If you randomly poke at some neurons, possibly altering their delay time or adding/disabling connections, statistically the longer the loop the more likely it is to be affected. This view also nicely explains the other effects of LSD. Your sense of time is wrong for the exact same reason your perception is distorted: the timing of all the loops have been slightly detuned.
    • fluoridation 4y ago
      What strikes me as unconvincing about interpreting vision as operating on the frequency domain is that anyone can look as a sound wave's spectrogram and understand what it's representing, even if they're totally unfamiliar with graphs in general. I've looked at spectrograms and noticed sounds that I was missing, then looked for them at the right times and frequencies and found them. You may not be able to understand the sound from a spectrogram (e.g. understand words, or melody), but you can understand what it's representing. If you look at the 2D FFT of a bitmap, the result is completely incomprehensible. You could look at two videos and their corresponding 2D spectrograms side-by-side and not know which corresponds to which.