This morning I went to see Christian Robert talk at QUT about the use of simulation as a tool in statistics (contains the slides from the talk). Overall it was quite an interesting talk although it was aimed at a less statistical audience than it was received by (mostly PhD students from Kerrie Mengersen’s BRAG group). I livetweeted the talk using the hashtags #xian and #amsi and have tried scraping my tweets using the tool described here. So I’ve now got a handy little archive of the tweets that I made. Turns out #xian is a poor choice of hashtag as it’s a region in China.
Edit: I’ve replaced the scraped tweets with a Storify thingo instead.

In a room full of nerds #xian #amsi0
likes·
0
comments 
Simulation: an ubiquitous tool for statistical computation #amsi #xian http://twitpic.com/alwhet0
likes·
0
comments 
Simulation useful for reproducing “chance” on a computer e.g. forecasting hurricane paths #amsi #xian0
likes·
0
comments 
Simulation also used to produce randomness in computer games such as World of Warcraft #xian #amsi0
likes·
0
comments 
Simulation can be used to approximate the area under a curve, i.e. an integral. This is taught in high school Maths. #xian #amsi0
likes·
0
comments 
Christian Robert giving us a teaser of his ABC talk later this afternoon. #xian #amsi0
likes·
0
comments 
Evolution of a population from a most recent common ancestor requires statistical inference for any meaningful analysis. #xian #amsi0
likes·
0
comments 
Not just the topology of the tree of life but the length of the branches are important for understanding how species are related #xian #amsi0
likes·
0
comments 
The uniform(0,1) pseudorandom generator is the building block of statistical simulation. Yes. The. #xian #amsi0
likes·
0
comments 
Intel working on true random generators but these may harm reproducibility and it’s not certain that they remain purely random. #xian #amsi0
likes·
0
comments 
The idea of finding pi by counting number of bivariate U(0,1) draws within the unit circle is the key to statistical simulation #xian #amsi0
likes·
0
comments 
Given a probability density f how can we produce randomness according to f which is fast but not based on approximation? #xian #amsi0
likes·
0
comments 
Generating dots within the brown region similar to the unit circle approach produces random values from f #xian #amsi http://twitpic.com/alwsub0
likes·
0
comments 
Acceptreject algorithm uses a function g which is a “hat” for f with a bounded ratio. Generate dots for g. #xian #amsi0
likes·
0
comments 
If f is complex we may not be able to find a hat g. Slice sampling uses random walk based only on values of f ratherthan density #xian #amsi0
likes·
0
comments 
Slice sampling does random jumps in vert and horiz directions to discover the shape of f, making sure we don’t jump outside. #xian #amsi0
likes·
0
comments 
MetropolisHastings useful where slice sampler fails. Explores locally to understand global picture, like in Civilisation V #xian #amsi0
likes·
0
comments 
Using standard Maths techniques to find minimum on a lumpy surface can be very difficult. Stochastic methods explore whole space #xian #amsi0
likes·
0
comments 
Travelling salesman problem requires minimising use of a resource over a huge number of possibilities. NPcomplete. #xian #amsi0
likes·
0
comments 
Simulation can be used to randomly perturb salesman circuit and accept changes probabilistically. #xian #amsi0
likes·
0
comments 
Christian Robert wraps up his talk with an illustration of using Metropolis algorithms to solve Sudoku problems fairly easily #xian #amsi0
likes·
0
comments 
Question about what makes a good random generator. Entire conferences dedicated to the topic, battery of tests to pass #xian #amsi0
likes·
0
comments