Post by mtriplett on Nov 28, 2020 2:01:54 GMT
Step 1: Introduction
I would like to introduce you to this topic by pondering this fundamental quote...
"Artificial Intelligence is the art of deciding what to do when you don't know what to do." - Sebastian Thrun
At some point, our old math stopped being useful to solve our modern problems. A few decades ago, some early pioneers recognized this issue, stopped, acknowledged that our old math didn't have all the answers, and began to come up with new ways forward.
Neural nets are one of those paths forward, a path to decide what to do, when we have no idea what to do.
Old School Math: Formulas
When most of us were growing up and learning math or science, we learned formulas for things. We calculated something by providing some values (lets say "x") and solving for some unknown value (we'll call it "y")
In a simple way, most of what we learned could be boiled down to:
y = f(x)
In plain English, "Y is some function of X." We can solve for any variable given the other variables and the function. Yes, we might have to juggle it a little with some algebra, but we rarely had to derive the function directly from nature. In school, they gave us lots of formulas, things like:
distance = rate * time
Now please consider for a moment:
Neural nets address this fundamental issue.
New School Math: Neural Nets
Neural Nets are just another tool. People talk a lot about their structure, weights, bias, gradient descent, etc. We'll get to that another time. It all sounds complicated and it is. However, the purpose is simple. "Neural nets derive functions when we don't know how to."
Think again about our generic function: y = f(x)
Neural Nets simply solve for the function (f), given the data (x and y). Because of this, neural nets are called "Function Approximators". They are not exact, but they can get very close. They don't produce a formula that we can read, but they can be used like a formula...plug in the inputs, calculate, get some outputs. It is important to realize that a neural net can approximate ANY function.
We can use them for useful purposes, without understanding what the formula is or how it was derived. Neural nets can be used to solve many previously unsolvable problems on their own without human help. To me, that is why they are important.
I hope you found this to be a useful first step. If not don't worry, every next step will be different.
Regards,
Martin
I would like to introduce you to this topic by pondering this fundamental quote...
"Artificial Intelligence is the art of deciding what to do when you don't know what to do." - Sebastian Thrun
At some point, our old math stopped being useful to solve our modern problems. A few decades ago, some early pioneers recognized this issue, stopped, acknowledged that our old math didn't have all the answers, and began to come up with new ways forward.
Neural nets are one of those paths forward, a path to decide what to do, when we have no idea what to do.
Old School Math: Formulas
When most of us were growing up and learning math or science, we learned formulas for things. We calculated something by providing some values (lets say "x") and solving for some unknown value (we'll call it "y")
In a simple way, most of what we learned could be boiled down to:
y = f(x)
In plain English, "Y is some function of X." We can solve for any variable given the other variables and the function. Yes, we might have to juggle it a little with some algebra, but we rarely had to derive the function directly from nature. In school, they gave us lots of formulas, things like:
distance = rate * time
Now please consider for a moment:
- What if you don't know the formula for something or how to derive a formula?
- What if it is not even possible to come up with a formula for something?
Neural nets address this fundamental issue.
New School Math: Neural Nets
Neural Nets are just another tool. People talk a lot about their structure, weights, bias, gradient descent, etc. We'll get to that another time. It all sounds complicated and it is. However, the purpose is simple. "Neural nets derive functions when we don't know how to."
Think again about our generic function: y = f(x)
Neural Nets simply solve for the function (f), given the data (x and y). Because of this, neural nets are called "Function Approximators". They are not exact, but they can get very close. They don't produce a formula that we can read, but they can be used like a formula...plug in the inputs, calculate, get some outputs. It is important to realize that a neural net can approximate ANY function.
We can use them for useful purposes, without understanding what the formula is or how it was derived. Neural nets can be used to solve many previously unsolvable problems on their own without human help. To me, that is why they are important.
I hope you found this to be a useful first step. If not don't worry, every next step will be different.
Regards,
Martin