10.07.2015 Views

Information Theory, Inference, and Learning ... - Inference Group

Information Theory, Inference, and Learning ... - Inference Group

Information Theory, Inference, and Learning ... - Inference Group

SHOW MORE
SHOW LESS
  • No tags were found...

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Copyright Cambridge University Press 2003. On-screen viewing permitted. Printing not permitted. http://www.cambridge.org/0521642981You can buy this book for 30 pounds or $50. See http://www.inference.phy.cam.ac.uk/mackay/itila/ for links.286 20 — An Example <strong>Inference</strong> Task: ClusteringInitialization. Set K means {m (k) } to r<strong>and</strong>om values.Algorithm 20.2. The K-meansclustering algorithm.Assignment step. Each data point n is assigned to the nearest mean.We denote our guess for the cluster k (n) that the point x (n) belongsto by ˆk (n) .ˆk (n) = argmin{d(m (k) , x (n) )}. (20.3)kAn alternative, equivalent representation of this assignment ofpoints to clusters is given by ‘responsibilities’, which are indicatorvariables r (n)k. In the assignment step, we set r(n)kto one if mean kis the closest mean to datapoint x (n) ; otherwise r (n)kis zero.{r (n) 1 ifk=ˆk(n) = k0 if ˆk(n) ≠ k.(20.4)What about ties? – We don’t expect two means to be exactly thesame distance from a data point, but if a tie does happen, ˆk (n) isset to the smallest of the winning {k}.Update step. The model parameters, the means, are adjusted to matchthe sample means of the data points that they are responsible for.m (k) =∑nr (n)kx(n)where R (k) is the total responsibility of mean k,R (k) (20.5)R (k) = ∑ nr (n)k . (20.6)What about means with no responsibilities? – If R (k) = 0, then weleave the mean m (k) where it is.Repeat the assignment step <strong>and</strong> update step until the assignmentsdo not change.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!