Correlating that with a database of real versus bogus signatures seems like basic lookup or a DB query,
Because the data point you want to find is most certainly not in the database.
So, how do you formulate an SQL query like this: "split this spectrum of measurement into a vector of 100 numbers, search the database for similar vectors" (what is similar?)
Point is: you can't. As you would need to know already what similar means, so first item in the vector is similar if it is +/- 0.5% variation. And the last item in the vector is similar if it is +/- 400 (absolut).
Databases for stuff like this are called vector databases.
A simple example is the load curve (electricity) of a particular industry, lets pick a bakery.
Make them identical except for one thing: one is baking twice as much bread, and has a second branch on the other side of the village to sell it. Call them "Mr. Small" and "Mr. Big".
So, the load curve of both bakeries will be identical. The second branch of "Mr. Big" has its own meter, so it is not included in the load curve of the bakery itself.
The load curves, usually a 15 minute interval measure, aka 96 measure points per day: are more or less the same. They start in the morning around 3:30 warming up the ovens. Continue with making various droughts, filling the ovens around 4:15 with the first charges ... both run a shop in front of the baking hall.
So, Mr. Small is an early bird and starts everything 20 minutes more early than Mr. Big.
Mr. Big has a double the size of baking hall with twice the amount of ovens.
Bottom line Mr. Big uses exactly twice as much power/energy.
If you normalize the load curves so, that the peak is "100%" (abstracting away the actual amount) and print them on thin paper: they look identical. With some minour point: as Mr. Small starts earlier, if you place the two papers on top of each other, the curves do not match. You have to move Mr. Bigs load curve to the left to match with the early start of Mr. Small.
And then: the load curves are identical, except for random noise of a freezer jumping on, or different lights in the shopping area, or the randomness of wat kind of coffee customers order.
AI and vector databases are optimized to figure such similarities.
BTW: such load curves are actually standardized - they are obviously called "standard load profile", because: it is for every bakery the same profile, only the total amount of energy varies.
So, we do not need AI to figure them out. But assume you have data which you never were able to "standardize" into "standard profiles" ... AI is good into looking at a million vectors of similar length ... but not necessarily the same length. With similar curve, but not necessarily a similar metering at every data point. Remember: Mr. Big has at every data point the double amount of Mr. Small, as long as they are baking, and not only selling in the afternoon.
AI is good in defining its own "standard profiles", and telling you into which "category" aka "standard profile" your current data set matches.
