One of my favorite things about SRECon is watching how the industry is marching ever closer to constrained Kalman filters and I’m going to seem like a prophet before long.
-
-
I'm a washed-up electrical engineer and I want to hear more 
-
Right so my 30 second rant: many tools are built based on the specific functionality of a piece of software or hardware or protocol, using the rules of how that specific thing works to detect/diagnose
-
But with good, generalizable models, and sufficiently high data rates, we are getting to the point where we can start modeling systems with a level or two of abstraction away, and then describing overall “system” behavior in terms of input/ouput
-
Imagine trying to monitor how a motor in a factory behaves to diagnose a problem with that motor. You can put together specific tools to monitor that motor based on how that motor works, right?
-
But that’s very specific. What if you could monitor the factory’s output and diagnose the failure in the motor with that data? You won’t need to model the motor specifically.
-
Iiiiinteresting. We're just at the cusp of using something similar in security to detect adversarial behavior (screw complicated ML models, all we need are some good linear regressions to find breadth-first & depth-first behavior; scanning and extraction of data).
-
I'm also minded of something I was hearing 10 years ago from EEs—"at high enough clock rate, digital signals problems become analog again"
-
Not sure if this is on the same wavelength, but all-ears to hear more
End of conversation
New conversation -
-
-
Do you mean something like a proof of universal linearizability, or the ability to build such a beast in practice?
-
Just the ability to build a sufficiently accurate model
-
Linearizability is not even strictly necessary, and I’m not even convinced that we couldn’t revolutionize the monitoring world right now using unscented Kalman filters
-
Gotcha. So this seems (to someone that had never previously heard of Kalman filters) to be a new application of the techniques used by quants to build trading strategies.
-
It comes from a similar area, yeah. It’s not exactly the same. A closer analog is embedded control, like flight controllers
-
Yep. The seemingly now-defunct
@VerdandeTech was working on that problem some years ago, albeit from an angle they labeled "case-based reasoning".
End of conversation
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.