4

I was wondering what the difference between ground truth and baseline is? Is it necessary that a system should always be tested based on ground truth or it can be interchangeably validated using a baseline?

lonesome
  • 173
  • @Tim But there is nothing about baseline and in fact, my question is quite different – lonesome Jan 29 '16 at 10:39
  • 1
    Can you provide some context for this? It's hard to see where you're coming from here. These aren't really that related. – gung - Reinstate Monica Jan 29 '16 at 13:43
  • On first reading I thought this was clear enough but based on your comments I agree with @gung. You need to specify in the question what you think baseline means. – John Jan 29 '16 at 14:18

1 Answers1

4

Both terms are big fuss words that describe pretty simple things and often are used inconsistently. "Ground truth" is the

data that is "known" to be correct

and "baseline" is

initial measurement data collected prior to intervention

In real life setting, you compare your model to the data that is available to you, that is most trustworthy of what you can get, and that is relevant for your purpose. In many cases you do not have data that is "known to be correct", but you have one that is just available. If you have several different data sources, than you are lucky and you can use all of them as reference while having in mind what are the differences between them.

Tim
  • 138,066
  • Do you mean these "different available sources" are baselines? What if we have one source of data but different techniques to apply to this dataset. What is baseline here? – lonesome Jan 29 '16 at 12:40
  • What do you mean by baseline? – Tim Jan 29 '16 at 12:43
  • I used to think that to see the effectiveness of a model, we apply the proposed model on a set of data and at the same time, we apply a known technique to the same set of data. The technique in my mind was baseline. Am I completely out of line? – lonesome Jan 29 '16 at 12:48
  • What is the "known technique" that you mention? – Tim Jan 29 '16 at 12:50
  • state-of-the-art? or some technique that is currently in use? For example, consider how Google answers to the queries. The known technique is what Google is using. Then imagine how Bing and Yahoo treat users' queries. Now, at some point of time, someone brings a new technique for same matter. Is this correct to say that all those "known technique" that Bing and Yahoo and Google are using are baseline? Is it correct to analyze our new technique against them by this assumption? – lonesome Jan 29 '16 at 12:53
  • You can use them, but such usage of "baseline measurement" is inconsistent with the definition. It seems that you are relating to some specific professional slang, so it is hard to answer your question accordingly. – Tim Jan 29 '16 at 13:10