Link by Lorien Pratt

Author:Lorien Pratt [Pratt, Lorien]
Language: eng
Format: epub
ISBN: 9781787696556
Publisher: Emerald Publishing Limited
Published: 2019-07-29T04:00:00+00:00

Source: Rick Ladd.

STATISTICS

Using evidence to make good decisions was traditionally the realm of the field of statistics. Today, statistics, AI, ML, and DI are deeply interwoven. Many ML techniques are based on statistical principles. Bayesian statistics are an important part of DI.

In general, statistical methods differ from AI by their focus on smaller data sets and their use of statistical assumptions as to the distribution of data. For instance, a statistical method might assume that the distribution of heights in a population follows a normal – or bell-shaped – curve. Using this assumption, a statistical method could draw reliable conclusions about an animal’s height given its weight from only a few examples of “height, weight” pairs of that animal.

In contrast, ML is often used for problems for which not as much can be assumed, and so it depends on a larger amount of data as a substitute for knowledge about data distribution. Since “Big Data” is now available in many arenas, this is today practical in a way that it wasn’t in the past.

Another significant difference between statistical methods and AI is that AI is usually used retrospectively, meaning that a data set is gathered in advance of any experimental design, and then AI is used to identify patterns in that data. Statistical methods can be used proactively as well, such as when designing a Randomized Controlled Trial (RCT) to test the efficacy of a new medicine.

Relative to DI, most statistical methods provide single-link answers, just as ML does, as was explained in Chapter 1. So, a typical statistical question might be to address a single-link question like “If I know a 10-year-old child’s IQ, how will that translate into their salary at age 25?” In contrast, DI concerns itself with what’s the best action to take, given this answer. For instance, given that I know the relationship between IQ and salary, what are the consequences of various choices I may make about raising that child, such as the choice of their school?

Finally, statistics does not typically concern itself with the combination of expert knowledge and data. A typical decision model is based both on human expertise (for links where data are not available), and on links where data are available.5

A statistical subfield that deserves special mention in this context is Bayesian statistics – an old field that has been experiencing a resurgence in recent years [146]. The basic idea of Bayesian statistics is this: what is the likelihood that you have cancer, given that you are a woman over 50. We don’t calculate the overall probability of cancer, but rather we calculate probabilities of sub-populations like this one. Another example: what’s the likelihood of your revenue growing next year given that your competitors don’t launch a new marketing campaign. Again, we’re asking not the overall probability but the probability conditional on some other situation holding.

Bayesian statistics is relevant to DI because there’s a close relationship between propagating causation in the world and propagating probability (indeed, we might think of

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.