What problems can solve the problems of TensorFlow?

Question

The TensorFlow homepage describes its purpose as a "numerical computation software library." Looking through problems with samples, it seems that the problem is always formulated as follows:

• Enter
• Model Parameters
• Required conclusion

Given some training data for 1) and 3), 2) can be calculated.

I see how this can be used to create bots, self-employed cars, image classifiers, etc.

Given the broad definition of "numerical computation," is there a lack of a class of other problems, what can it be used for? Can this be used, for example, for more classical numerical calculations, such as air flow around an airplane or deformation of a structure under stress? Do you have any examples of how these classic problems should be formulated to fit the form above?

Answer 1

A good discussion of what artificial neural networks could do, the fact that our brain is a neural network may mean that ultimately an artificial neural network can do the same tasks.

A few more examples of artificial neural networks used today: making music , location based on images ,, google voice , stock trading forecasts , nasa star classification , traffic management
Some fields that I know about but do not have a good link for:
Optical quantum mechanics test setup generator
medical diagnosis, safety link only
Sharp LogiCook microwave, wiki , nasa mention

I think that there are many millions of "problems" that can be solved with ANN, a solution for presenting data (input, output) will be a problem for some of them. Some useful and useless examples that I thought of:
- A home thermostat that studies your wishes with certain types of weather.
- forecasting the production of bakery products - recognize stones on the board and display their locations - guessing personal activity and turning on the appropriate device.
- Recognize a person based on mouse movement.

Given the correct data and network, these examples will work. Dad has a computer that controls the heating system at home, I trained a network based on his 10-year data on heating (outside the pace, inside the pace, humidity, etc.), Unfortunately, I am not allowed to connect it.

My aunt and uncle have a bakery based on six-year sales data, I trained a network that predicts how much bread and rolls they should make. He showed me how important the correct source data is. At first I used the day of the year, but when I switched to the day of the week, I saw a 15% increase in accuracy.

I am currently working on a network that will detect go board in this image and display all 361 places telling me if there is a black, white or not a given stone.

Two examples that showed me how much information can be stored in one neuron and different ways of presenting data: An example of an image , an example of neurons (unfortunately, you need to prepare both examples yourself to give them a little time.)

For your example, airflow around an airplane.
I don’t know anything about airflow calculations, and my attempt will be a really huge three-dimensional input layer, where you can "draw" the plane and the direction and speed of the airflow.
This may work, but it requires tremendous processing power, someone who knows more about this particular topic probably knows a more abstract way of presenting data that leads to a more manageable network. This nasa document talks about a neural network for calculating airflow around a wing. Unfortunately, I do not understand which input they use, maybe this is more clear to you.