Maybe our dreams are (nothing but) Fourier transformations of daily events!
It was 1993 when I left my PhD thesis at the Fine Arts Faculty of Bilkent University unfinished. I was planning to study the effects of a global computer network and its possible impact on cultural and artistic products, like making millions of free copies of every song, film, game and book and distributing them freely via this network. How should people pay for these free copies? That was my main concern. Then I saw the alpha version of the Mosaic browser later that same year and immediately decided to drop my PhD thesis: A world wide web was already happening.
Then in 1994 I moved to the USA, and attempted a second PhD in a new subject, on my lifelong passion: hacking the brain. I then proposed,
- “Learning, thinking and human to human communication happens via packets of cultural patterns called memes”.
While thinking on this problem (1), the same day I hit upon another idea (2) and its unexpected ramification (3):
2. All five of our senses create signals for the brain and these signals, which are actually patterns (memes) that exist in cartesian (x, y, z, t) world format, become frequency (1/t) domain signals after being processed by the brain.
My first reaction was:
3. Oh, our dreams are actually Fourier transformations of the daily life! This reaction was possible since I was an electrical engineer by education and I had taken courses about signal processing and Fourier transformations.
Background: Years ago, when I was younger, a friend of mine had given me LSD as a practical joke and I had had a terrible day after that experience. It was, what some people call a bad trip. Then, after sobering up, it had become obvious that my mind, under the influence of the chemical, was behaving very similar to dreaming, except that I was wide awake and walking. There was little or no sense of time, similar to when I was dreaming. It was as if the time axis of the signals were non-existent and everything was happening in frequency domain. Events far apart in time looked similar if they had a similar frequency. A random guy I meet one day may look like my dear uncle who passed away long ago and could trigger past memories of him. (I will explain this later).
We do not have a model of the brain…
I was well aware that we humans did not have a model of the brain: How it operated, and how thinking, dreams, remembering and learning occurred inside our heads were all mystery. This awareness was the result of reading articles and books by neuroscientists like Karl Pribram and physicists like David Bohm. My sister who was a brilliant medical doctor, had also played a big part in this awareness. Yet, I was constantly surprised to see tons of articles and books (and lots of films) which were written as if we know how the brain functions.
At the time, I had read several books about holographic brains, holographic universes and similar memes but those books usually ended up with new-age ideas I did not find fully convincing. However, something in those books attracted my interest: I had seen many 3D images of toys and things recorded onto glass plates using laser light beams. When they were broken or divided, pieces of these plates displayed the same 3D image only with less resolution. The smaller the piece, the fuzzier it looked. I was aware that these pictures, made with mono-frequency laser light resulted in frequency domain recordings of 3D spatial objects. And the image was recorded all over the plate using a form of Fourier transformation: the recording was not local to any part of the plate but distributed all over it.
These holographic examples gave me a vision for some model of the brain but still many pieces were missing.
Then comes a book: “On Intelligence”
I had to leave my second PhD thesis unfinished also, since my department chair at the Ohio State University explicitly stated that I will not be honored a degree if I pursued my weird ideas. Luckily, I had a phone call the same week about a well paying job back home and returned soon to build the first web pages of Turkey.
Some years after that, in 2004, I saw a book on the pages of amazon.com and ordered a hard copy the same day: Written by Jeff Hawkins, it was called “On Intelligence”. Devouring it in a matter of hours, I had a shocking realization: This book was a revelation!
First it was bravely stating the obvious but occluded fact that we humans still lack a model for the workings of the brain. And second, it was heroically attempting to introduce a model. My greatest takeaway from the book, at least initially, was to learn about a magnificent result by a scientist called Vernon Mountcastle.
Mountcastle showed that, the important part of the brain, called the neocortex (or the cortex) was composed of the same type of neurons and processed incoming sense signals with one algorithm. On page 50, Jeff Hawkins wrote: “He proposes that the cortex uses the same computational tool to accomplish everything it does.” I was flabbergasted!
With my friends, genius programmer Hakan Güleryüz and math wizard (today a brilliant university math professor) Ferit Öztürk, we had been experimenting with Wavelet methods and neural network algorithms to solve various signal processing problems around 1998 and 1999. They were for clients who wanted, for example, to find a specific carpet design among 500, within seconds in a factory’s dim lighting conditions.
We immediately understood the implications of the important result we found in the Hawkins book. All human senses hear/see/feel various kind of signals but all signals become one and the same kind of electrochemical signal once inside the cortex, while traveling on axons and dendrites. Our experience with wavelet programming gave us some understanding of how signals could have been interpreted in the cortex. Wavelets are a kind of finite Fourier signals and are used to solve pattern recognition and various similarity problems.
Around the same time we had come across a few Siggraph (a conference on computer graphics) papers on image recognition and recalling, which employed wavelets and special functions called the “Haar functions”. The terminology these papers used were “multi resolution analysis”. Haar functions were very simple functions used for wavelet calculations: you only used addition and temporal shifting. One Siggraph paper around 1995 or 1996 explained how to recall a given image from a crowded database using very fast multi resolution analysis of images using Haar wavelets. I am not able to find that great paper now but here is a more recent example for doing a similar job.
But how is that related with the cortex?
How are the signals entering the Block Box called the cortex related with the Haar wavelets? And how do these signals become the same type?
Sequential signals (signals with x, y, z and time) flow on neural channels and they get divided at junctions… Weights are formed… Bu weightler de akan action potların miktarı ile orantılı oluşur. Patterns and sequences…
All signals entering the cortex i/o cells get stored inside a vast array of neural jungle as the same kind of information, and there is only one known mechanism for the signals carry information towards the cortical cells: Action potentials. Action potentials are electrically loaded proteins…
The answer to these questions are very simple! In a Haar wavelet, which is itself a form of Fourier transformation on finite signals, only two operations are permitted (add and not-add), one can imagine electrochemical signals can be simulated with such an algorithm: Synaptic operations are designed to either pass or not-pass the electrochemical signals, nothing more complicated than a flow of currents happen between neurons. Haar methods work on multiple layers of hierarchical data, hence the term “multi resolution analysis”.
Note: This atricle is not yet finished! I plan to complete it soon.
