reading classics (#2)

The second “classic” this year was Gelfand & Smith’s 1990 paper on the Gibbs sampler, Sampling-based approaches to calculating marginal densities, as presented by my Master student Guillaume Revillon:

I just went back and checked at the corresponding post of last year, on the same paper. The drawbacks of the presentation are rather similar, namely staying too close to the paper (like adopting the bracketing convention!) and missing a critical analysis of the functional approach adopted by Tanner & Wong and reported by Gelfand & Smith. However, I think Guillaume fared better (than Xiaolin last year) when asked to explain points to the class (which had never heard of Gibbs sampling for most of them!) and by running his own illustrative simulation.

At a more personal level, I felt more keenly how the early days were focussed on the functional perspective (as opposed to the Markovian one). For instance, the paper used m replicas at each Gibbs iteration to provide a more accurate of the distribution of the chain at this iteration and very few Gibbs steps, like 4 or 10 (in the examples presented by Guillaume). After the presentation I thus commented to the class that this was somehow misleading in that more complex and realistic models needed time (and iterations) to reach areas of interest.

One Response to “reading classics (#2)”

  1. Dan Simpson Says:

    This is really interesting! It’s great to see how the first wave of practitioners thought this would all pan out.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.