Boffins build neural networks fashioned out of DNA molecules Scientists have built neural networks from DNA molecules that can recognise handwritten numbers, a common task in deep learning, according to a paper published in Nature on Wednesday. It’s pretty easy for traditional convolutional neural networks. Now, scientists are testing wackier models on the MNIST database of training …
It's "computation" in the theoretic, Computer Science definition of the term.
Basically it maps the "image" (by hand) into a piece of DNA.
Run it next to a bunch of other pieces of DNA
Filter it with a process that deactivates less matching DNA and leaves the more matching DNA active.
In principle it could run a very (1000s, millions?) of class classification test in the same time it runs a few. So I guess you could brute force the Traveling Salesman problem (maybe), by enumerating every variation of short paths, then combining every variation of every variation of those short paths to get every variation of the entire full network path, then run them all through selection.
But TBH I think the DNA/RNA/Ribosome is already Turing complete, given it can read a "tape" (the DNA), write a tape (I'm pretty sure ribosomes can be made to generate new DNA strands) and carry out actions based on the contents of that tape.
Carnegie Mellow were working on chemistry based logic gates (with IBM funding) decades ago, using dyes for signal I/O. It never seemed to occur to anyone that what was really needed was molecular sized electron injection and extraction. Photons are 100s of times bigger than most molecules.
I'm sure one of the later Ender books (Children of the Mind?) features an alien civilisation who communicate by sending digitally encoded DNA sequences, which the receivers are supposed to 'read' by incorporating them into their own DNA.
Tangent, but that's what this reminded me of.
The Register - Independent news and views for the tech community. Part of Situation Publishing
Biting the hand that feeds IT © 1998–2022
