if you don’t know the definition of “interesting” a priori: context mixing, probably
If you have a signal where a certain portion of points are noise, and a certain portion of points are "interesting", what's the "best" way to determine which points are "interesting" as they come in over time?
-
-
-
i.e. run a ton of hypothetical algorithms looking for “interesting” items based on different possible parameters, and weighted-mix the results of the ones that have found the best fit so far according to some measure designed to exclude noise
- Još 9 drugih odgovora
Novi razgovor -
-
-
Check for inequality with the least interesting number.
-
is there really such a thing as a least interesting number?
- Još 1 odgovor
Novi razgovor -
-
-
SNR - or ratio of signal strength to noise - for radio signals fft comes to mind, kalman filter for estimation, or stdev for deriving SNR from data ( eg: Bollinger band, moving avg ... ). Often depends on specifics though
- Još 3 druga odgovora
Novi razgovor -
-
It’s out of my character to say this, but some machine learning categorizer is probably the thing. This is something CERN has been doing at the LHC site from the start - they produce way too much data to process it all so they need to reject most data as likely-uninteresting.
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
Without a heuristic for 'interesting', noise is indistinguishable from signal. Machine learning is just a process for iteratively deriving a model for this heuristic, but you can also do it with other a priori models for 'interesting.' (without the drawbacks of machine learning.)
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
can do something like this https://en.m.wikipedia.org/wiki/Grubbs%27s_test_for_outliers … , or just use a “robust” estimator over the data
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
Determine a metric that exists in interesting data that ideally does not exist in noise data then test for it. that's best and ideal But that usually doesn't happen so you aim for one that's exclusive a high % of the time then add metrics to refine until the threshold is good
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.