ML bubble?

Two Sigma / Eng
shhhhhhhh_

Go to company page Two Sigma Eng

 shhhhhhhh_
Jul 2, 2018 23 Comments

Feel like people are greatly overestimating the usefulness of machine learning. It isn't needed in the vast majority of cases, and when it is useful logistic regression does the job fine 99% of the time.

Also, people are paying big bucks for people with stats degrees who can't get a useful model into production. I think software engineers with basic knowledge of ML are a lot more useful than data scientists with basic knowledge of software engineering.

Am I crazy or is it only a matter of time before ML cools off and becomes a super oversaturated field?

23

comments

Want to comment? LOG IN or SIGN UP
  1. TOP
  2. RECENT
  3. OLD
 TOP 23 Comments
  • New / Eng
    level5.99

    New Eng

    PRE
    Amazon
     level5.99
    You guys have no idea how deep learning transformed the tech industry in certain areas like CV, NLP, speech, dialogue etc. Only in hedgefund world like TwoSigma simpler ml models are used, actually most hedgefunds only use linear ml models...... which are way different and actually 'simpler' than tech companies. That is why you have this bias. That being said, most people today in tech industry only uses ML, not really develop/do research on ML.
    Jul 2, 2018 2
    • Two Sigma / Eng
      shhhhhhhh_

      Go to company page Two Sigma Eng

       shhhhhhhh_
      OP
      I actually have no idea what models we use for our trading division. This bias is mainly comes from my previous FANG job.
      Jul 2, 2018
    • New / Eng
      FGAN

      New Eng

      PRE
      Google, Facebook
       FGAN
      I had an interviewer at TS talk about building NLP models (sentiment), and I saw some openings for deep learning roles. What team/org are you in?
      Jul 2, 2018
  • Uber / Other
    TheJuice

    Go to company page Uber Other

     TheJuice
    Not crazy. Way too much hype.
    Jul 2, 2018 0
  • Microsoft / Product
    @zzz

    Go to company page Microsoft Product

     @zzz
    If you look at about 5 years ago, cloud computing was being used like that. Everyone wanted to put something something cloud into whatever they were building. Was some of that pure bullshit and hype, yes, but doesn't mean it wasn't transformational. Think something similar is going with ml and deep learning. Large parts of what you hear everyday is just hype, people want to be relevant and jumping on the bandwagon, but doesn't mean it's not a transformational technology. Most people actually don't realize how much of ml based algorithms they perhaps use daily, weather, stocks, finance, news articles, search, recommender systems, mapping, complex optimizations such as matching your Uber driver, fraud detection, and the usual suspects in speech, vision, self driving cars etc.
    Jul 2, 2018 0
  • Microsoft / Eng
    Soylearnt

    Go to company page Microsoft Eng

    BIO
    TC - still 1 comma
     Soylearnt
    Agreed the ML is a bit overused term. However I have seen first hand how ML has helped solved some existing problems and saved resources and $. I have met too many folks yapping AI/ML but met few folks who understood what I want to solve and then told me if ML is the right solution or ... You got to apply some filter when you hear ML.
    Jul 2, 2018 2
    • Two Sigma / Eng
      shhhhhhhh_

      Go to company page Two Sigma Eng

       shhhhhhhh_
      OP
      It helps sometimes for sure, especially for recommendation systems. You don't need a stats PHD to implement collaborative filtering or logistic regression though.
      Jul 2, 2018
    • Microsoft / Eng
      Soylearnt

      Go to company page Microsoft Eng

      BIO
      TC - still 1 comma
       Soylearnt
      This is my observation at Microsoft. Folks with formal education like PhD usually are the ones who are creative and come up with algo. Average software engineers implement those algo or tweak them to get results. I am sure there are good ML engineers who have knowledge as a PhD person. Anyway the trick is to identify one that has the chops to apply ML and hit the ground running. This is where we have differences between data scientist and data engineers.
      Jul 2, 2018
  • I work on a ML platform. Yes, there's a lot of hype and it expectations will fall soon.

    What needs to die right away is all the people who think AGI is right around the corner. No, we are no where close to machines replacing any complex task.
    Jul 2, 2018 1