2

If deep learning is a black box, then why are companies still investing in it?

nbro
  • 39,006
  • 12
  • 98
  • 176
Case Msee
  • 129
  • 2
  • 1
    Companies invest because deep learning has been very effective for many companies in achieving desired goals. The 'black box' neural network itself is the easy part. The hard part is deciding what inputs to use, what representation should those inputs be presented in, what should be the expected outputs and representation, what utility functions will achieve the best results and in many cases collecting a set of training data that will actually lead to good results. That's what they are investing in. I disagree there are only small numbers of systems doing something real. – Dunk Jan 02 '19 at 18:39

2 Answers2

5

Easy answer: utility.

The strength and applicability of "black box" NNs has been regularly validated in the past few years, and business is concerned with results. (i.e. they don't care how the sausage is made, so long as it gets made.)

nbro
  • 39,006
  • 12
  • 98
  • 176
DukeZhou
  • 6,237
  • 5
  • 25
  • 53
  • :Very useful answer. One of my friend ask me the question about iPhone facial recognition .Can we define this process to lay person?How things are happening at backend to tag my photos?How iPhone facial recognition works.Any mathematical evidance?.How can we inspire someone who is machine learning enthusiast and needs some facts to enter into this research.I need to motivate my junior researchers with useful answers – Case Msee Jan 01 '19 at 03:49
  • Yes, I think you're right, but I think it would be also nice to at least mention that, in certain areas, like healthcare, people are reluctant to use deep learning (or ANNs) because of its "black box" nature. See also explainable AI. – nbro Aug 20 '19 at 23:20
2

I think that the universal approximation theorem plays a large role in why companies and governments are investing in deep learning, it states that theoretically an ann can approximate any continuous function with n-dimension input variables. Specifically it states that feed forward nets with a single hidden layer can do this lending credence to the implication that rnns and cnns are also capable of universal function approximation. So they are investing because they have continuous functions that need to be approximated and really the best tool for the job is neural networks.

nickw
  • 317
  • 1
  • 7
  • 3
    RNNs have been proven to be Turing complete and CNNs have also been proven to be universal approximators, so the powerfulness of RNNs and CNNs is not just hypothetical. See https://ai.stackexchange.com/a/13319/2444. Anyway, I don't think companies are investing in neural networks because they are theoretically powerful. AIXI is also theoretically an optimal agent, but nobody uses it. – nbro Aug 20 '19 at 23:22
  • 1
    Well, I think AIXI isn't used due to the fact that it is incomputable and the approximations of it being less effective than other comparable RL algorithms. – nickw Aug 21 '19 at 13:59