PhD Speaking Qualifier: "Non-parametric maximum likelihood clustering with Dark Cuts"

[Matt Barnes]

Hi everyone,
I have some interesting recent work to present for my PhD speaking
qualifier on Monday at 8am in GHC 7101 (it's the only time that worked for
everyone. Seriously). So wake up bright and early and let's talk about some
cool math.

Non-parametric maximum likelihood clustering with Dark Cuts

*Dec 7 (Monday) at 8:00 am GHC-7101 *
Speaker: Matt Barnes

*Abstract:* Clustering algorithms traditionally rely on assumptions about
the data distribution or use a heuristic for the objective function. I
propose a non-parametric clustering model and show that the maximum
likelihood estimator is equivalent to MinCut, plus an extra meaningful term
defined as the Dark Cut. The maximum likelihood estimator can be solved for
in polynomial time with strong guarantees using a randomized algorithm.
Preliminary experimental results show Dark Cuts performs comparably to the
best spectral clustering methods.
…
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.srv.cs.cmu.edu/pipermail/autonlab-research/attachments/20151204/9f4e1421/attachment.html>