[ArXiv] 3rd week, Feb. 2008
It seems like I omit papers deserving attentions from time to time. If you find one, please leave a message. Even better if a summary can be left for a separate posting.
Wavelet papers:
- [astro-ph:0802.2377] J. M. Lilly & S. C. Olhede
On the Design of Optimal Analytic Wavelets - [math.ST:0802.2424] Autin, Le Pennec & Tribouley
Thresholding methods to estimate the copula density
A statistics paper and astro-ph papers adopted statistical tools:
- [stat.ME:0802.2155] Guellil & Kernane
A New Approach of Point Estimation and its Application to Truncated Data Situations - [astro-ph:0802.2105] N. Padmanabhan et.al.
The real-space clustering of luminous red galaxies around z<0.6 quasars in the Sloan Digital Sky Survey - [astro-ph:0802.2446] Banerjee & Ghosh
Evolution of Compact-Binary Populations in Globular Clusters: A Boltzmann Study II. Introducing Stochasticity - [astro-ph:0802.2944] E. W. Rosolowsky et.al.
Structural Analysis of Molecular Clouds: Dendrograms - [astro-ph:0802.3185] G. Efstathiou
Limitations of Bayesian Evidence Applied to Cosmology - [astro-ph:0802.3199] A. A. Mahabal et. al.
Automated Probabilistic Classification of Transients and Variables
Simon Vaughan:
Have any of the Astrostats guys been following the discussion of the value of Bayesian methods in cosmology that’s going on in the UK? The latest part is George Efstathiou paper mentioned above, but other papers include those from Andrew Liddle (arXiv:astro-ph/0701113) and the exchange that followed. Very interested to know what the opinion is of these. My knowledge of Bayesian methods is rather basic, which perhaps explains why I don’t quite see what Efstathiou’s argument really is. Sure, Bayesian model selection is dependent on priors – however you perform the calculations – but we all know this. But non-Bayesian methods are dependent on often implicit choices such as choice of statistic etc. I suspect there’s a more subtle argument but I don’t get it and wondered if anyone here can explain it to me…
02-29-2008, 4:33 pmBTW: Any comments on the use of the word ‘evidence’ by the ‘Cambridge’ school? Seems like almost the entire world (esp. Bayesian stats community) uses the word differently, but I guess its firmly established in the UK cosmology community.
vlk:
I am not a Cosmologist, so I am very likely missing most of the Efstathiou’s argument. But we had a very nice talk by Rocky Kolb here yesterday, and he made a point that is perhaps relevant to this controversy. He pointed out that “dark energy” is an inference made by comparing a cosmological model to the data, and that you can avoid appealing to dark energy by considering any number of alternate models. In other words, your calculations are only as good as the model (or set of models) you have choosen to look at. So I don’t really get why Efstathiou seems surprised.
02-29-2008, 6:10 pmhlee:
Not knowing both Bayesian and cosmology, but only reading a few Bayesian evidence related papers from the statistics side as a part of model selection and a couple of Andrew Liddle’s papers (I liked his introductory cosmology book which led me into reading his papers), I cannot answer to your question to the satisfactory level. Despite my ignorance, my feeling is that all matters with how cosmologists set up (probabilistic) models and their believes in those models. I rather call them assumptions but their meanings are lost in translation between statistics and cosmology. I understood “evidence” as “degree of penalization” and penalization heavily depends on model assumptions. Nonetheless, I admire cosmologists’ broad minds of adopting various statistics. Over the months, topics in cosmology utilized statistics appeared most frequently in arxiv compared to other astronomy subjects. I do appreciate your comment very much, which motivates me to dig in this subject and might force me to write (or to ask someone to write) about “Bayesian evidence” to drag more attention from others.
03-03-2008, 12:25 pm