[HN Gopher] Generalized Sequential Probability Ratio Test for Fa...
___________________________________________________________________
Generalized Sequential Probability Ratio Test for Families of
Hypotheses [pdf]
Author : luu
Score : 18 points
Date : 2026-02-20 07:59 UTC (3 days ago)
(HTM) web link (sites.stat.columbia.edu)
(TXT) w3m dump (sites.stat.columbia.edu)
| measurablefunc wrote:
| Great. How do I use this in my life to make things better?
| glial wrote:
| The SPRT is probably already making your life better: it's used
| to decrease the cost of medical trails, optimize
| classifications in high-stakes examinations (i.e. for medical
| certifications), detect defective manufacturing processes, etc.
| It sounds like this paper extends the method to groups of
| hypotheses, whereas the basic version is limited to a null
| hypothesis and an alternative hypothesis.
| data-ottawa wrote:
| This helps with determining when have you observed enough data
| to make a decision.
|
| A/B tests, monitoring metrics, health, quality control all use
| this.
|
| If you use LLMs, you might use this to determine if a model
| update or prompt change impacts results using fewer tokens.
| noelwelsh wrote:
| From 2014, with 43 cites in Google Scholar:
| https://scholar.google.com/scholar?cites=4354840623154369813...
|
| Might want to check the cites for more recent work ("Efficiency
| in sequential testing" looks relevant) and also the literature on
| "bandit best arm identification", which seems to be distinct from
| this line of work but tackles broadly the same problem.
___________________________________________________________________
(page generated 2026-02-23 23:00 UTC)