Finding regions of high values? I have an array of a million float values, and I seek to identify regions of abnormally high values.
For instance, within [0.39, 0. 41, 0.4, 0.67, 0.7, 0.8, 0.37, 0.49], [0.67, 0.7, 0.8] constitutes a region of abnormally high values.
Are there standard algorithms/methods for addressing this type of problem? I was thinking about filtering for values that exceed say 2 standard deviations from the mean, but such a method does not take into account how these regions are preferably connected. Furthermore, 2 standard deviations is just an arbitrary threshold I conjured up.
Maybe we could borrow a clustering method from machine learning?
 A: Since no one has answered it I'll post my crude approach.

Furthermore, 2 standard deviations is just an arbitrary threshold I conjured up.

Well at some point you will have to choose the threshold. Instead of choosing it by SD though, I'd pick a percentile (ie. $5\%$ of the most extreme values). Choosing a percentile based threshold does not require you to make any assumptions about distribution, independence, etc. 
All you'd need to do is order your data, and select the value corresponding to the highest 5% of the data $X_h$ (ie. approximately $5\%$ of data is higher than $X_h$). Then any data points higher than $X_h$ can be considered 'abnormally high'.
Preferring areas of connected sets is also something you'll have to make a decision on. One approach would be to choose a sufficiently large $k$ such that the probably of having $k$ abnormally large values in a row is small but still likely to occur often given you have $1,000,000$ values. This $k$ will depend on your choice of percentile.
For example, if $k=4$ you'd expect this to happen around $6-7$ times for $1,000,000$ values (very crude calculation, I didn't bother making it exact since there's a bunch of stuff to consider) for a $5\%$ percentile.
So any $k \geq 5$ is so unlikely to occur multiple times it's not really worth you considering. This means you can run an algorithm $4$ times to find instances where abnormally large values occur $1,2,3,$ and $4$ values in a row. This should not take long to run for a dataset of just $1,000,000$ values, even using for() loops and such.
