-2
$\begingroup$

Usecase - Gmail detecting spam format emails is a simple NLP system based on some clear rules. Doesn't rely on artifical neural network. No training involved.

Usecase - Gmail predicting auto-completion of sentences is again a simple NLP system based on some grammar. Doesn't rely on artifical neural network. No training involved

Usecase - Opening bank account online is a simple expert system, which doesn't rely on neural network. Training is involved as part of inference phase.

++++

What kind of problem, that triggers a need to approach artificial neural network based system? in the below active areas...

enter image description here

Note: Industry is very excited to use neural network based AI, with lack of clarity on this aspect

Improve this question
$\endgroup$
6
  • 1
    $\begingroup$ Why do you expect there to be a specific answer. The only correct answer is "when neural networks give better performance". $\endgroup$ Commented Sep 26 at 21:22
  • $\begingroup$ @Digitallis I just added the active areas, to make it little specific $\endgroup$ Commented Oct 2 at 19:39
  • $\begingroup$ How is "opening a bank account" an expert system !!!???? $\endgroup$ Commented Oct 2 at 20:09
  • $\begingroup$ @Digitallis datascience.stackexchange.com/questions/134459/… $\endgroup$ Commented Oct 2 at 20:26
  • $\begingroup$ @Digitallis I wasn't really sure what that referred to, either. If you're classifying everything into "neural network" or "expert system" buckets, rule induction approaches used in anti-fraud systems would fall into the latter category, as would the online "open an account!" interactive flowchart system (mimicking the behaviour of a bank customer-interface person). $\endgroup$ Commented Oct 5 at 13:59

1 Answer 1

Reset to default
6
$\begingroup$

Your question is broad, so I'll have to also answer broadly. Based on my own experience in industry, I generally use the following set of guidelines:

  1. if a good rules based solution exists to a problem, and either (i) you don't expect the rules to change, or (ii) manually/semi-dynamically updating the rules is feasible, then a neural network is probably not necessary
  2. if the inputs and outputs are relatively simple and interpretable, then a simpler ML model (linear regression, decision trees, random forest, etc) should probably be tried first because they are just as effective as NNs, but easier to maintain
  3. if a simpler model isn't immediately obvious, and relationships between input variables exist but the extent of their correlation is unclear and there is high dimensionality and a lot of data to train on, then it may be worth turning to a neural network. this is often at the expense of interpretability
  4. a few common use cases: neural networks are the go-to for image classification (especially CNNs), and transformer architectures for NLP problems related to text prediction are two that come to mind
Improve this answer
$\endgroup$
4
  • $\begingroup$ what does CNN stand for? $\endgroup$ Commented Sep 29 at 18:53
  • 1
    $\begingroup$ convolutional neural network $\endgroup$ Commented Sep 29 at 19:02
  • $\begingroup$ which category does Agentic AI & Generative AI fall under? among this four point list.... $\endgroup$ Commented Oct 2 at 19:04
  • 2
    $\begingroup$ the categories i wrote out don't cover everything, but Agentic and GenAI are broad and come from NLPs and LLMs for text based analysis, and image related ML models for stuff like image classifcation, generation, etc $\endgroup$ Commented Oct 2 at 20:30

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.