11
aviophile
14d

I am learning Machine Learning via Matlab ML Onramp.

It seems to me that ML is;
1. (%80) preprocessing the ugly data so you can process data.
2. (%20) Creating models via algorithms you memorize from somewhere else, that has accuracy of %20 aswell.
3. (%100) Flaunting around like ML is second coming of Jesus and you are the harbinger of a new era.

Comments
  • 6
    Add "scouring the internet for examples and finding 4 year old medium blog posts that use an outdated tensorflow version"
  • 1
    Or rolling out your own ML.
    Most of the time spent on ML is preparing your dataset for training, and then n-fold validation to get that extra 0.34% over other algo....
  • 3
    Or is it a genuine field of study that has massive applications and is used by a whole lot of people to do a lot of stuff that just can't be done classically (and also to make good money in the process)? I wonder.
  • 1
    @RememberMe To me, current machine learning beats 5 year old child by its sheer speed, which it owes to semiconductor revolution. But what do I know?
  • 3
    @aviophile whut? How about recommendation engines, for example? Ad targeting, market analytics, object detection and recognition, automatic machine translation, OCR, {denoise, supersampling, and upscaling} operators that beat classical ones, speech recognition and synthesis, face recognition and identity systems, chatbots, conversational agents and assistants, camera optics improvements, machine and device design, and so on?

    These are all kinds of deployed ML systems btw, and they work and make big money. I'm not even talking about research, which is doing some pretty insane stuff.

    DL models already beat humans at many things and can be scaled effectively, and replicating the human brain is *not* a goal of standard ML/DL, that's computational neuroscience.
  • 0
    Second coming of what?
    God is dead and so is his son.
  • 0
    If you think the preprocessing is 80% now, just wait until you start developing custom models with unique goals; tunning and debugging are 95% and preprocessing 5%

    I am currently working on a project where this has been the case for almost a month— preprocessing part completed in a few days.
  • 0
    @amoux by completed, I meant automated.
Add Comment