Event: Wrong Models, Bad Stats and the dimensional Stochastic Bayesian
Date: 28 February 2020
Time: 12:00
Venue: Lab 602, Maths & Physics Building
Presenter: Prof. Lino Sant
"Models are Wrong but useful" has achieved mantra status amongst modellers. It needs consideration and clarification.
This talk will start by mentioning past successes in mathematical modelling and the many dualities they created within academic output. Time and randomness were confronted vigorously through dynamical systems and probability theory more than a century ago. Eventually, the various strands merged into the study of stochastic processes.
Statistics provided the link to orientation and formulation for practical use. Powerful as the methods developed were, controversies over their use and interpretation raged furiously. We look at current ones which emerged within the statistics and applied quantitative researchers communities. The validity of quoting p-values has been hotly contested. The scourge of lack of reproducibility has robbed published empirical research of its credibility in many sciences. However, the re-emergence of the Bayesian paradigm having achieved sophistication and widespread use, offers hope.
In its last section the talk will explain how the ultimate flight into infinite dimensional analysis, first effected within deterministic systems again more than a century and a half ago, has also achieved maturity within the stochastic setup for quite a while. The synthesis with Bayesian methods was particularly fertile and wide ranging in scope within the modelling agenda being requested by various areas of research and other productive activities.
Examples taken from research work done by the author illustrating the issues will be presented during the talk.