9 Distribution shift

 

Wavetel is a leading (fictional) mobile telephony provider in India that expanded its operations to several East and Central African countries in recent years. One of its profit centers in the African markets is credit enabled by mobile money that it runs through partnerships with banks in each of the nations. The most straightforward application of mobile money is savings, first started in Kenya in 2007 under the name M-Pesa. With mobile money savings, customers can deposit, withdraw, and transfer funds electronically without a formal bank account, all through their mobile phones. (Remember that these transactions are one of the data sources that Unconditionally evaluated in chapter 4.) More advanced financial services such as credit and insurance later emerged. In these advanced services, the bank takes on financial risk and can’t just hand out accounts without an application process and some amount of due diligence.

9.1      Epistemic uncertainty in machine learning

9.2      Distribution shift is a form of epistemic uncertainty

9.2.1   The different types of distribution shift

9.2.2   Detecting distribution shift

9.2.3   Mitigating distribution shift

9.3      Adaptation

9.3.1   Prior probability shift

9.3.2   Covariate shift

9.3.3   Concept drift

9.4      Robustness

9.4.1   Prior probability shift

9.4.2   Covariate shift

9.4.3   Concept drift and other distribution shifts

9.5      Summary