this the video i watched

i get what he is saying its just weird to me that it works so well, doesnt seem very intuitive

phd in computational biology

Niceee what area do u work in? I work on sequencing informatics

oh s*** lol thats rly cool, like bioinformatics and working with DNA/RNA data? ik a lot of ppl who do that stuff

I do mainly image a***ysis/machine learning/statistics basically, like processing images and acquiring/analyzing data from them

Im hopefully starting a new project soon that will be more in the theoretical statistics/machine learning realm

Yea I work as a clinical bioinformatician, so processing/analyzing DNA/RNA sequencing data for cancer cases etc

It’s a lot of fun n very rewarding for sure (and the medical ppl are a lot cooler than CS ppl tbh 😹)

That image processing stuff is really sick tho, we of course work closely with pathology and some of the histopathological tech that’s coming out is pretty mind blowing

nice fam that sounds very cool, yea ive been recently thinking it would be nice to learn more about the sequence a***ysis tech.

Im not so familiar with the histopathological tech, tbh my lab is mostly developmental biologists so im often learning more about the experimental tools

That AI music is crazy

I have this weird idea, basically I'm thinking what if we make networks that instead of being feed forward and giving an output based on input, we make networks that use unsupervised techniques to generate clusters that individually specialize in learning particular parts of the data structure. The idea being that these smaller subnetworks learn from the data and share what they learn with other subnetworks, and in theory you could mix and match subneteorks for different tasks

It's basically a probabilistic graph network like a bayesian network

I wanna make a model that makes music videos

But how imma do it is a secret

I wonder if reinforcement learning would outperform standard CNN for classification of image data

He knew

Video editing & animation AI programs already exists

Almost everything you can think of in the creative space already have programs that are already available or deep in the process

Im excited for upscaling & repairing AI to get better for old media

yea i looked into it u right

turns out they saying i dont have enough time left in my phd to start a new machine learning project. me graduating is more important than trying to cram in more stuff because of funding issues

im not too upset, i might do a post doc in a ML lab, in the last year or so imma try to do some self study and maybe use this time to apply more ML to my current project

they said theres a new transformer architecture that incorporates fourier transform into the classification process

im like duh, the idea of transforming your data before feeding it to a neural network seems very obvious but apparently not so common