Hacker News Reader

Comment by badminton1 | Hacker News Reader

"A job in deep learning".

It is highly unlikely that you will get a job in which you exclusively use deep learning alone, and not any other ML/AI technique.

Once you learn DL, then, "congratulations... here are 100 other topics you might need to know about before getting a job". http://scikit-learn.org/stable/tutorial/machine_learning_map...

View on HN

I disagree. I think we might soon be at a point where someone might be able to get a job just knowing how to use CNNs well. Why do I think this? Well .. CNNs have basically licked the problem of image classification. They require a lot of trial and error. So .. I can totally see people packaging this up (e.g. NVidia's DIGITS or TStreamer), and CNN skills become sort of like Word/MS Office for some industrial applications. These people won't get paid 500K .. more like what web developers make. Just my personal opinion.

I think you're right that the barrier is a lot lower now that there is a somewhat successful blackbox abstraction.

In the past you need to know that to recognize lines: hough transform, recognize polygons: line simplification, recognize face: cascades, etc etc.

Now? You can almost just feed it arbitrary labeled training data and do well without any sort of feature engineering. Just another api to glue.

Of course there will be more packaged applications, but I don't see them being something as low level as a CNN. More like applications of them such as speech recognition, translation, motion control, etc. Things that you can actually treat as a black box and integrate them into a system.

Would someone in their right mind create a self driving car product with the help of someone who learned deep learning from a blog or youtube? probably not.

You're definitely right that most ML jobs aren't DL all day everyday. But if you are working with data that is rich and fairly homogeneous but hard to model (like recorded speech) then you'll probably benefit a ton if you use DL due to the ability to learn underlying representations. If that's what you are up to then you do spend all of your time building and tuning DNNs (all three ML people at Deepgram are doing that at this very second :) )

Reply on news.ycombinator.com