I later found out that biology is not, in fact, boring. But you can certainly teach it in the most boring manner possible.

Oh, no doubt about it. The proof that channels have a finite capacity was a huge breakthrough. It’s just odd that he hadn’t seen the formula before. He reinvented it and used it very differently

And here is p. 125 of my copy of Planck's "Wärmestrahlung" (1905). He even does "max entropy". But I think von Neumann had understood this more deeply, because by realizing it is the classical limit of the vN entropy, it was not at all limited to phase space probabilities, but any states, really.

This is the way.

You should also tell them that when Shannon came to him in 1946 or so to ask what he should call his p log p formula, he told him to call it entropy, but not telling him that he already wrote down that formula as the classical limit of the vN entropy in 1927. "It has been used under that name".

Brown nose by proxy

But they forget that every mutation can change the fitness effect of every other mutation via epistatic interactions. So, you don't need to invoke rapidly changing environments: every mutation coming after a beneficial mutation within the fixation time window can change the adaptive trajectory 2/2

That's basically the plot of "Satanic Verses" (the part about the puppet, not the AI).

Once you've done all this you can come back and use my info-theoretic tool, which will just extract the information. But you need to know information about *what* (i.e., annotated phenotype). That tool (IDSeq) does it better than any ML tool in existence, here or anywhere else in the universe.

But you can actually binarize the byte, and you can do the same with your abundances (turns out the organism only cares about high/low info). Here, the ten classes (numbers) will be immediately apparent, but if you know the phenotype on a subset, you can use that for training.

Here's a description of dimensionality reduction techniques for hand-written images. The images are both simpler and more complex than the microbiome: there are 782 "abundances", but each has a range of one byte. colah.github.io/posts/2014-1...

How many biomes? What do you want to predict if you knew the patterns?

Approximate cloning is totally allowed.

I have been complaining about this many times. I try to submit to journals with active editors (who are involved in the decision making and don’t just count balls and strikes). But with varying success.

This is because pain is a construct of our brain. A pain signal and a pleasure signal (say) look the same as far as the spike train goes. So you can do this if you can simulate the brain. Good luck with that.

Ummm.