QUT School of Mathematics Postgrad Day – Day 1

Every year the School of Mathematical Sciences (in whatever incarnation) holds a day (or two) of talks where the higher degree research students present what they’re working on. It’s a great chance for students to get some practice for presenting to conferences.

  1. Modelling an Inbound Passenger Facilitation System using an Object Oriented Dynamic Bayesian Network

  2. samclifford
    Jegar Pitchforth working on improving airport efficiency with Airports of the Future. Object Oriented Bayes Networks. #qutsmspg // @bjkraal
    Thu, Sep 06 2012 20:10:15
  3. samclifford
    Turning Rehbein framework into a Bayesian Network allowed the asking of deeper questions, making the metrics quantifiable #qutsmspg
    Thu, Sep 06 2012 20:15:48
  4. samclifford
    Validated Bayesian Networks based on psychometrics validity framework. Data can be tricky to obtain, esp. validation data #qutsmspg
    Thu, Sep 06 2012 20:20:19
  5. samclifford
    Long term goal for AotF includes an ecosystem of models, visualisation, validation. #qutsmspg
    Thu, Sep 06 2012 20:25:08
  6. Wayfinding in Airports: A Bayesian Network Approach

  7. samclifford
    Charisse Farr talking about BNs for AotF again. This time on wayfinding within the airport #qutsmspg
    Thu, Sep 06 2012 20:31:02
  8. samclifford
    Farr has very fancy slides. I think I’ve seen the package before but never used it. #qutsmspg
    Thu, Sep 06 2012 20:32:05
  9. samclifford
    Helping make decisions for figuring out whether to refurbish an airport baed on an increase in passenger numbers. #qutsmspg
    Thu, Sep 06 2012 20:33:04
  10. samclifford
    Wayfinding the consistent use and organisation of visual clues for navigating an environment. Not just sign placement #qutsmspg
    Thu, Sep 06 2012 20:34:34
  11. samclifford
    Theories of cognition very important in understanding human factors in wayfinding. Modelling with BNs allows quantitative analysis #qutsmspg
    Thu, Sep 06 2012 20:38:35
  12. bjkraal
    @samclifford We get our own models from that data, and we can inform other models, eg BNs.
    Thu, Sep 06 2012 20:42:48
  13. samclifford
    Human factors have a huge impact on the overall effectiveness of way finding. Mostly communication and air travel experience #qutsmspg
    Thu, Sep 06 2012 20:43:05
  14. nitrojane
    @samclifford my theories on cognition are “humm. Interesting. That is certainly….cognitive.”
    Thu, Sep 06 2012 20:43:15
  15. samclifford
    @nitrojane I an tell you spend a lot of time cogitating on this
    Thu, Sep 06 2012 20:43:49
  16. samclifford
    Communication of results also important. Not everyone knows stats. Scorecards, Excel summaries. And R/MATLAB. #qutsmspg
    Thu, Sep 06 2012 20:48:13
  17. Bayesian spatiotemporal modelling of chronic diseases and risk factor exploration

  18. samclifford
    Time for spatiotemporal modelling with Jannah Baker. Type II diabetes modelling #qutsmspg
    Thu, Sep 06 2012 20:50:54
  19. samclifford
    Baker outlining the systematic review of spatiotemporal disease modelling she did. Presented at ASC2012? #qutsmspg
    Thu, Sep 06 2012 20:53:31
  20. I had a talk to Baker’s supervisor afterwards and it’s largely what she presented at ASC2012 (which I didn’t get to attend as I had Healthy Buildings on at the time). Still, it was nice to see a review and to see where the project will go.
  21. samclifford
    Sparse Poisson MCAR models can be used to examine spatial correlation between types 1 and 2 diabetes #qutsmspg
    Thu, Sep 06 2012 20:58:30
  22. Novel Phase I-II Dose-Finding Design Methodology

  23. samclifford
    Darssan Balasingam talking about design of dose finding studies #qutsmspg
    Thu, Sep 06 2012 21:03:18
  24. samclifford
    Non-monotone dose efficacy. Is a logistic regression with a quadratic enough? surely a nonparametric function estimator #qutsmspg
    Thu, Sep 06 2012 21:16:00
  25. I’d like to add at this point that one of the other people in the room asked a question based on the same sort of thinking. How do we know that this will capture the non-monotonicity of the efficacy curve? Clinical trials aren’t really my thing in terms of statistics, but dose-outcome logistic regression is something I’ll have to familiarise myself with for some work on UPTECH.
  26. Bayesian hierarchical modelling of spatial and spatio-temporal variation in cancer outcomes

  27. samclifford
    Back to spatial stats with Susanna Cramb. Small area data for cancer incidence / cc @Eats_Food #qutsmspg
    Thu, Sep 06 2012 21:22:38
  28. samclifford
    Cramb’s slide explaining the CAR model has a brown 74 Mustang as background #qutsmspg / cc @bjkraal
    Thu, Sep 06 2012 21:24:11
  29. samclifford
    Spatial relative survival models quite rare but very neat. Local and global smoothing of excess deaths due to a disease #qutsmspg
    Thu, Sep 06 2012 21:26:30
  30. samclifford
    Stepwise inclusion of tumour stage and socioeconomic data for colorectal and breast cancers. #qutsmspg
    Thu, Sep 06 2012 21:29:19
  31. samclifford
    Spatial variation in breast cancer seems due largely to SES factors. Not so for colorectal where it’s mainly age #qutsmspg
    Thu, Sep 06 2012 21:30:47
  32. samclifford
    Remoteness plays a huge part but the spatial effect of breast cancer really comes down to access to services. #qutsmspg
    Thu, Sep 06 2012 21:31:30
  33. I’ve seen Cramb talk a few times now and her modelling is always that little bit cooler when I see it.
  34. Bayesian Inference for Soil Carbon in Farmscapes

  35. samclifford
    Ben Fitzpatrick talking about spatio-temporal modelling of soil organic carbon emissions. #qutsmspg
    Thu, Sep 06 2012 21:40:30
  36. samclifford
    Fitzpatrick and Cramb seem to be looking at the same model but with different approaches. #qutsmspg
    Thu, Sep 06 2012 21:43:18
  37. samclifford
    CAR model one way of getting around the anisotropy when looking at complex spatial data such as soil carbon. #qutsmspg
    Thu, Sep 06 2012 21:45:29
  38. samclifford
    Adjacency weights in CAR model based on empirical semivariogram. I’d just go for an anisotropic covariance function #qutsmspg
    Thu, Sep 06 2012 21:48:56
  39. samclifford
    Suggestion that anisotropy might be taken into account by fitting covariates and bilinear spatial trends then look at residuals #qutsmspg
    Thu, Sep 06 2012 21:50:41
  40. Fitzpatrick’s only in his first year (iirc) of his studies but his project looks quite interesting. I raised with him that anisotropic covariance functions exist for Gaussian processes (he referenced Banerjee’s book, so I thought he might be aware) but it sounds like he’s going to try to work on using the discrete CAR model. I think a continuous space model might be more effective for looking at the total organic carbon emissions from farm soil because unless you’re interpolating based on your CAR model (which is “bad” as far as I’m concerned) you can’t really estimate the farm emissions based on discrete points.
  41. Informative priors for Bayesian spatial models, with application to cone-beam CT scans

  42. samclifford
    Matt Moores rounding off the session with using fan beam CT scans as informative priors for cone beam CT treatment planning #qutsmspg
    Thu, Sep 06 2012 21:54:34
  43. samclifford
    External field prior for Potts model drastically improves tissue identification in cone beam CT based on fan beam diagnostic #qutsmspg
    Thu, Sep 06 2012 22:06:38
  44. The impact of spatial scales on the outcome of Bayesian spatial models

  45. samclifford
    Su Yun Kang talking about the impact of scale in spatial modelling. Cancer incidence in Qld again but with point data. #qutsmspg
    Thu, Sep 06 2012 22:42:00
  46. samclifford
    I really enjoy watching Su Yun talk. She does cool stuff with random processes and we were in an NPBayes group to learn this stuff #qutsmspg
    Thu, Sep 06 2012 22:45:19
  47. samclifford
    And she uses INLA for fitting her spatial point process models. #qutsmspg
    Thu, Sep 06 2012 22:46:32
  48. samclifford
    Interesting to use the chi-square test for model comparison. Why not the probability integral transform? #qutsmspg
    Thu, Sep 06 2012 22:52:00
  49. Kang’s talk was really interesting because it was work I hadn’t seen from her previously. Looks promising.
  50. Bayesian statistical models for understanding health-related outcomes for women screened for breast cancer

  51. samclifford
    Jeff Hsieh talking about spatial modelling for breast cancer screening. I wonder how similar this is to Nicole White’s. #qutsmspg
    Thu, Sep 06 2012 23:02:53
  52. samclifford
    So I’m watching this talk about breast cancer screening and the impact of access. And I remember that LNP are messing it up. #qutsmspg
    Thu, Sep 06 2012 23:04:50
  53. samclifford
    And for rural women, access is already difficult. How can we save women’s lives if we make it even harder to serve them? #qutsmspg
    Thu, Sep 06 2012 23:05:38
  54. samclifford
    It makes me so mad to think that a cancer that is such a prevalent killer is having its prevention gutted. #qutsmspg
    Thu, Sep 06 2012 23:07:56
  55. samclifford
    Relative risk of breast cancer mortality increases as you go more rural even as you include SES and other geographic factors. #qutsmspg
    Thu, Sep 06 2012 23:12:25
  56. The Queensland LNP government have recently been taking a razor to the public service and the programs they deliver. Breastscreen Queensland is being completely restructured and decentralised and there have been about 3000 jobs shed in Qld Health today as a result of the Newman government’s insistence that the public service is too big. So hearing about cuts to services while watching a presentation talking about the effect of access on survival rates is pretty heartbreaking.

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 )

Twitter picture

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

Facebook photo

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

Google+ photo

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

Connecting to %s