Search Mailing List Archives

Limit search to: Subject & Body Subject Author
Sort by: Reverse Sort
Limit to: All This Week Last Week This Month Last Month
Select Date Range     through    

[theory-seminar] Theory Lunch 08/12: Mingda Qiao

Prasanna Ramakrishnan pras1712 at
Thu Aug 12 11:56:45 PDT 2021

Reminder: Theory lunch starts in a few minutes, with the talk starting half
an hour later! See you all there!

On Mon, Aug 9, 2021 at 1:09 PM David Wajc <wajc at> wrote:

> Hi all,
> This week's theory lunch will take place Thursday at noon (PDT), at this
> zoom room
> <>. As
> usual, we'll start with some socializing, followed by a talk starting at
> 12:30pm. Pras will be MCing. Thanks Pras!
> Mingda will tell us about: *Exponential Weights Algorithms for Selective
> Learning*
> *Abstract:* We study the selective learning problem, in which a learner
> observes $n$ labeled data points one at a time. At a time of its choosing,
> the learner picks a window length $w$ and a hypothesis $h$ from a given
> hypothesis class $H$, and then labels the next $w$ data points using $h$.
> The excess risk incurred by the learner is defined as the difference
> between the average loss of $h$ over the $w$ data points and the smallest
> possible average loss among all hypotheses in $H$ over those data points.
> We give an improved algorithm, termed the hybrid exponential weights
> algorithm, that achieves an expected excess risk of $O((\log\log|H| +
> \log\log n)/\log n)$. This gives a doubly exponential improvement in the
> dependence on $|H|$ over the best known upper bound. We also prove an
> almost matching lower bound, showing that the $\log\log|H|$ dependence is
> necessary.
> Based on joint work with Greg Valiant.
> Cheers, David
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <>

More information about the theory-seminar mailing list