/Tech2h ago

Online-To-Batch Reductions Generate Frequency-Scaled Confidence Intervals For Learners

2211297

Original post

But if you are running an online to batch reduction, you can get valid hints by forming confidence intervals around conditional means for points you see frequently in your training sample. The width of the interval hint now scales naturally with the frequency of the point.

Aaron Roth@Aaroth

But what if your online learner had a hint: Every day t, it received an interval [a_t,b_t] and the promise that the true mean was in the interval. Now it can learn optimally and randomize only within the interval. But where can you get a hint from? Online it would be impossible.

6:50 AM · Jun 19, 2026 · 128 Views

Sentiment

Sentiment building, check back later.

Cluster Engagement

Digg Deeper

No Digg Deeper questions have been answered for this story yet.

Posts from X

Most Activity

VIEWS128LIKES1

Aaron Roth@Aaroth

So do this. You get a randomized predictor whose feature conditional variance scales smoothly with the frequency of x. And once you have this you have avoided the obstruction in the red/blue example above, and fixing the randomness works!

Aaron Roth@Aaroth

2h12810

BOOKMARKS1RETWEETS1REPLIES1

Aaron Roth@Aaroth

This predictor is perfectly calibrated, but no deterministic predictor with range {1/3, 2/3} can be calibrated here, so fixing randomness fails. If you think about it for a bit, the problem is the combination of high variance of prediction and high weight in the distribution.

2h4111