[LOCKSUIT]

I didn't really grasp Bayesian but now I do. Is the below what Bayesian is?



Physicians intuitively use Bayesian statistics on a daily, if not hourly, basis. Here’s why:

When a patient presents with a symptom, such as chest pain, the physician considers the possible causes (etiology) of that symptom in a rank-ordered list, from most likely to least likely. This rank-ordered list is referred to as a differential diagnosis for the presenting symptom(s).

https://www.aafp.org/afp/2005/11...

The physician then asks a series of probing questions meant to re-rank that list of potential etiologies, such as “do you experience chest pain primarily when climbing stairs or exercising?” The answer to each successive question re-ranks and in general narrows the list of candidate etiologies.

The re-ranking of diagnoses based on each successive question asked by the physician is premised on the predictive power of multiple discrete facts over the predictive power of fewer facts.

Inference engines that incorporate Bayesian statistics include backwards-chaining (inferential) expert systems, and are among the examples dating back to the 1970s of the application of artificial intelligence methods to medicine.

[AndyGoode]

I believe Bayesian just refers to Bayes' Theorem (also called Bayes' Rule):

https://en.wikipedia.org/wiki/Bayes%27_theorem

It's a simple formula you can probably memorize in a matter of seconds. The applied AI technique they mention is called Bayesian networks:

https://en.wikipedia.org/wiki/Bayesian_network
https://www.norsys.com/tutorials/netica/secA/tut_A1.htm

Those are called that name because they are based on Bayes' Theorem. The reasons Bayes' Theorem isn't used more in applied AI, especially in expert systems, is that it gets very complicated very fast, and often some of the probabilities needed are not available. It wasn't until computers got faster in the early 2000s that Bayesian networks could be run at all for AI applications, so it's pretty clear that human brains don't use Bayes' Theorem for making probabilistic estimations.

[LOCKSUIT]

That is same idea though, my OP explains it concisely in full.

Seeing it now, I think: It was so simple this whole time. The way it was presented to me wasn't right.

[goaty]

I believe Bayesian just refers to Bayes' Theorem (also called Bayes' Rule):

https://en.wikipedia.org/wiki/Bayes%27_theorem

It's a simple formula you can probably memorize in a matter of seconds. The applied AI technique they mention is called Bayesian networks:

https://en.wikipedia.org/wiki/Bayesian_network
https://www.norsys.com/tutorials/netica/secA/tut_A1.htm

Those are called that name because they are based on Bayes' Theorem. The reasons Bayes' Theorem isn't used more in applied AI, especially in expert systems, is that it gets very complicated very fast, and often some of the probabilities needed are not available. It wasn't until computers got faster that Bayesian networks could be run at all for AI applications, so it's pretty clear that human brains don't use Bayes' Theorem for making probabilistic estimations.

Sorry to disagree again,  Bayes models are only as complicated as you want them to be, simple solutions suffice for simple tasks, the same as expert systems.

[AndyGoode]

Sorry to disagree again,  Bayes models are only as complicated as you want them to be, simple solutions suffice for simple tasks, the same as expert systems.

Let me reword my statement: Bayesian nets have been around a long time but the computation time becomes exponential as the size of the problem grows because the number of needed nodes in a Bayesian net becomes exponential, so Bayesian nets were used only rarely for sizable nets until the early 2000s, when computers became fast enough that substantial networks could be run.

A graphical model specifies a complete joint probability distribution (JPD) over all the variables. Given the JPD, we can answer all possible inference queries by marginalization (summing out over irrelevant variables), as illustrated in the introduction. However, the JPD has size O(2^n), where n is the number of nodes, and we have assumed each node can have 2 states. Hence summing over the JPD takes exponential time.

https://www.cs.ubc.ca/~murphyk/Bayes/bnintro.html

But, building Bayesian models have been criticized in the past as a very time consuming effort which is computationally expensive.

https://www.quora.com/What-are-some-reasons-for-the-increasing-popularity-of-Bayesian-inference

[LOCKSUIT]

But what I said in my opening post is bayesian....Someone prove me wrong....

And what are these large bayesian doctor expert machines? Huh? You have a list of diseases, narrow them down, then narrow those down? Or, narrow down the narrowdowning voters themselves?

[HS]

So Bayesian is common sense based on knowledge? Like the scientific method is common sense based on mystery? Its how we think anyways, but they've just made it formal so we are able to stay on track.

[LOCKSUIT]

I say being formal is really a waste of time, this idea was so simple all along. Confusion is actually bad explanations.

[AndyGoode]

But what I said in my opening post is bayesian....Someone prove me wrong....

And what are these large bayesian doctor expert machines? Huh? You have a list of diseases, narrow them down, then narrow those down? Or, narrow down the narrowdowning voters themselves?

By the way, I had to review Bayesian nets to refresh my memory to answer your question accurately, since I've never liked them, so I rarely study them or think about them on my own.

Anyway, I think they're used mostly as a cause-and-effect diagram, where all possible effects that might interest you are listed at the end of the network, which is usually at the bottom of the diagram, with associated probabilities for each possible outcome. For input you tell it what is happening, then the probabilities of each possible effect appear immediately at the bottom of the graph. I think H.S. got it right: they're like a common sense machine where all the math is done so quickly in the background that you don't have to think about how the answer you got was derived. I was going to say they're like a simulation, but that's not very accurate, and H.S.'s explanation is much more accurate. They're not like a simulation very much because you have to know beforehand everything that might happen for the Bayesian network, and you have to build all such consequences into the network at the bottom, which is a lot of work, from what everybody says, whereas in a simulation you can just watch what happens visually without needing to list possible outcomes or to concern yourself with probabilities.

Some applications I've seen...

1. health
inputs = diet, exercise
outputs = probabilities of having high blood pressure or chest pain

2. wet ground
inputs = rain event, car wash event
outputs = probability of slipping on wet ground

3. trout fishing
inputs = location, fish description
outputs = probability of having caught each of the species listed

4. alarm event
inputs = burglary event, earthquake event
outputs = who will call you if the alarm goes off

Machine Learning | Bayesian Belief Network
Published on Aug 13, 2019
RANJI RAJ


Bayesian Networks
Published on Mar 25, 2015
Bert Huang

[LOCKSUIT]

I'm sure GPT-2 can answer what entails. Nividea is onto chatbots for health etc already. We've already been testing GPT-2. GPT-2 can do it. Rain>slip. Look:


I went out into the rain to get my car key
. As I was walking up the ramp on the opposite side of the gate, I hit my head on a railing, causing a concussion. The next thing I know, I'm being carried from the gate on my stomach. My glasses and belt come off the gate. I manage to grab a beer and pull myself to the side of the building. In the ambulance, I was told I had a broken nose and a brain injury.

The doctors told me the damage was likely to last up to a year. So I'll need to be in rehab for a while. My recovery was difficult to deal with. I was unable to do much.

While in recovery — the days got worse, I couldn't go anywhere without getting a headache. I tried to go to the gym and play football with the guys around me but I couldn't because I couldn't get up. I'm not too tough to have a cold, but it became a recurring soreness that persisted.

[WriterOfMinds]

I feel like I've seen people use "Bayesian" in popular speech as a description of a vague/general behavior pattern or even a life philosophy, which comes down to, "I will update my current ideas or plans whenever I receive new information" (i.e. avoid rigid thinking) and "I will do my best to assess the probability of things accurately."

In technical terms, however, "Bayesian" means exactly what Andy says.  Bayes' Theorem is a piece of statistical math that lets you calculate the probability that Thing A will be true if you already know Thing B is true and know some things about "conditional probabilities" (correlations) between A and B.  A Bayesian process would be one that uses Bayes' Theorem to determine knowledge or make decisions.

The description in the OP is consistent with how a Bayesian system would behave, but it does not tell the whole story.