Could someone explain in simple terms:

• what machine learning really is?
• how it will be used in msfs?

Re machine learning I have only really come across it as part of an advertising campaign (like my last iPhone) so I don’t really understand what it really is. It does, however, seem to be one of the things that makes this sim genuinely next generation so I’m interested to know what it will do for us. 

An example...Microsoft has taken satellite imagery of a specific area and taught a machine to identify where a tree/building should be based on the image. The machine then creates a tree/building with same dimensions based on the imagery and place it on that spot. Therefore it learned how to create a tree/building in the sim without someone actually having to create a tree or building on their own. Which is apparently how most of the scenery was created for the entire globe where imagery exists. Something that would probably take a coder many many years. 

The basic idea of machine learning is that you can get some software to do a specific task without explicitly programming how to do that task.

To take an example that applies to MSFS: finding the outline of a building in some satellite imagery.

First you need a training set of data, so in this case it would be the vector data of plenty of known building outlines and the satellite images used to draw them by hand.
You then start with an algorithm that has a certain set of inputs. In this simplified case it would be image processing settings relating to detecting edges, noise reduction, drawing best fit vectors etc.
Finally you have a way of comparing the actual output to the desired output from the training data.

Let this run over many iterations and the algorithm will optimise itself to perform the required task.

Once the algorithm has optimised itself, it can be let loose on the full data set.

(I'm simplifying a lot here, before anyone tears into this 😀)

Thanks,  very useful. 
 

so do you think part of the tech alpha could be about giving testers specifics regions to fly around and identify where the ‘machine’ has made errors so they can go back and improve the training manual? Just thinking of the observation that one of the journalists made about parts of Cairo being decorated in sand coloured trees because it had got confused about what it should be drawing there. 
 

i find this technology fascinating. I also imagine then that the sim will keep improving over time as well? One of the devs was talking about how the sim will slowly change over time and that they have already seen changes to the french coastline through climate change over the development period. 

1 hour ago, Superdelphinus said:

so do you think part of the tech alpha could be about giving testers specifics regions to fly around and identify where the ‘machine’ has made errors so they can go back and improve the training manual? Just thinking of the observation that one of the journalists made about parts of Cairo being decorated in sand coloured trees because it had got confused about what it should be drawing there. 

It seems to me that the machine learning is being used for placing trees and vegetation, finding bodies of water and rendering autogen buildings where photogrammetry isn't available.

There's no way that they can check the entire earth to make sure the ML algorithms are performing well in all cases, so getting more eyeballs on it will definitely help.

In the Cairo example, the ML algorithm is maybe trying to evaluate what colour to make the tree so it blends in well with its surroundings. The optimised algorithm being used possibly worked well for temperate climates but not for desert.