I’ve been looking at life in terms of systems bc that’s the realm of my knowledge. But life has been reminiscent of an autoencoder lately.
Last year (for me) was all about the encoder. Compressing a high amount of information/reducing dimensionality to find the minimal sufficient representations. Like in linear algebra where you take a high dimensional system and project it to a (stable) lower subspace.
We all know that with any compression there is information loss. What is determined as important is biased on what it actually deems as important. And with overcompression you may lose a lot of important information that you didn’t even realize was that important.
But at the same time compression is important bc it takes a high dimensional space and reduces it in order to build stability in what may have been unstable (due to the sheer amount of info). It is needed. This is the bottleneck.
And once the latent space of minimal representations become stabilized the decoder can then come in and begin reconstructing the structure while discarding anything that was considered as “noise”. It reconstructs based on what it has learned. What it has understood. The generalization process. Reintegrating. Now it can start to expand again.
