AIRG,
Today Xiaomin Zhang will be talking to us about robust regression. This
particular approach achieves a consistent estimator in the face of
adversarial data corruption. Moreover, a variant based on gradient
descent is fast and scalable. (They claim 20x speedup compared to the
best L1 solver.)
4pm, CS 3310
https://papers.nips.cc/paper/6806-consistent-robust-regression
https://arxiv.org/abs/1506.02428
(While I normally prefer the canonical version from the publisher, in
this case I think the arXiv version is easier to read.)
Aubrey
On 4/15/19 11:08 PM, XIAOMIN ZHANG via AIRG wrote:
> Hi AIRG,
>
> I will talk about robust regression this Wednesday, which does
> much better than standard regression at handling noisy data, corrupted
> data, or outliers. Robust regression via convex methods may not be
> consistent. However, non-convex methods can be consistent. I will focus
> on hard thresholding for robust regression.
>
> References are the following papers:
> 1. https://arxiv.org/pdf/1506.02428.pdf
> 2. https://papers.nips.cc/paper/6806-consistent-robust-regression.pdf
>
> See you at 4pm Wednesday, CS 4310.,
> Xiaomin
>
>
> _______________________________________________
> AIRG mailing list
> AIRG@xxxxxxxxxxx
> https://lists.cs.wisc.edu/mailman/listinfo/airg
>
|