What if universe is just a crypto algorithm?
What if universe is just a crypto algorithm that tries to settle itself by increasing it’s entropy to be the most unique in its way.
If you have read two of my previous articles, you know about multi perspective neural network (MPNN) algorithm.
Let’s recap: MPNN is an algorithm that gets the input data and applies convolution operations to it so that each layer coming out of convolution layer will be compared with each other by calculating cosine distance of each pair in the convolution layers. Cosine distances are pushed to be as much as possible so that we ensure each layer in convolution is different from each other. This operation is applied several times so that a small embedding is acquired when the convolution layers are concatenated at the end. This embedding will be unique for that image and similar to the embeddings that belong to the similar images if we retrain our network for each image individually.
So, how can this phenomena can be used?
We can create NFTs that contain unique tokens for the images so that when a similar image or the same but a part of it is owned by someone else, an IP protection claim could be done.
On the other hand, this can be used also for self supervised learning.
Imagine, two similar images are fed into the MPNN and unique embeddings are generated. We then calculate log-sum-exp of the error and calculate gradients for the “new" embedding. We check which columns have the least amount of gradient. If they are small, it means that they are unique features that define that concept and should be kept.
On the other hand, two different images are fed into the MPNN and unique embeddings are generated. We then calculate log-sum-exp of the error and calculate gradients for the “new" embedding. We check which columns have the big amount of gradient. If they are big, it means that those are features that should be kept because they explain the reason od difference.
By applying these steps, we can filter our embedding from unnecessary and misleading information.
Okay, recap finished.
Now imagine, the universe is just this algorithm. Because the cosine distance is pushed to be maximized, the entropy of the information in the world increases. Each time passing, more information is created and all this information is being pushed away from each other. There will be a one moment where this universe has enough entropy amount so that it will stop growing. It will just stop. Time will stop everything will stop and basically freeze. And we won’t even understand and feel wtf is happening. Does a floating point in a vector feel what is happening with it when the vector stop changing? No.
I will share the code for two use cases that I mentioned above in the next article.