29 Day 29 (May 5)

29.1 Announcements

  • Mathematical statistics workshop (link)

  • Teaching evaluations

  • Calendar

    • Lecture today Tuesday 5/5
    • Lecture on Thursday 5/7 (if we don’t finish today)
    • Presentation dates/time? My preference for presentations is May 4-8. Please send me an email () and provide three 30 min long times/dates that work for you.

  • New overview stats paper on physics informed neural networks (link)

    • Mechanistic vs. descriptive models
    • Physics informed neural networks really blends the two

  • Questions/clarifications from journals

    • “What is INLA? The paper mentions INLA as a common approach for fitting log-Gaussian Cox process models, but it does not fully explain how INLA works. I understand that the paper focuses more on showing how LGCPs can be fitted using mgcv, but I still need to understand what INLA is and why it is commonly used for these spatial models.”
    • “In particular, I would like to better understand what criteria are used to determine whether a model adequately captures the spatial structure and how this approach can be combined with other validation techniques to make reliable modeling decisions.” See Tilman Gneiting’ work here and here.
    • “I am trying to understand the functional data analysis (FDA), which is an approach used to analyze and model data that varies continuously over time or space. I am reading about it and it seems that is very used in Plant Pathology. Could you tell me more about this technique?”
    • “Still hard to understand/explain in my own words, what mechanistic models could do better than descriptive methods.”
    • “My question for today is why should we use a mechanistic spatio-temporal model instead of a traditional spatio-temporal model? I understand that traditional models can describe relationships between variables, but I am still trying to understand when they become too limited.”
    • Modeling space and time as seperable in descriptive models

29.2 Mechanistic spatio-temporal models