I was working on the mnist and pets notebooks yesterday when I got a call from a Republican friend running a small software company down here in Southern California. I was lamenting the aparent politicization of the Covid response - it generally seems Republicans want the response to weigh more heavily toward the economy while Democrats want it to weigh more heavily toward preservation of life.
As I was just focussing on the details of binary cross entropy and loss functions in general, that got me thinking, what is the loss function we should be trying to minimize in a pandemic? Surely it’s not loss of life exclusively, since we make decisions all the time that contemplate some loss of life, presumably in order to obtain some other benefit (like driving cars, for example). And surely it’s not the economy regardless of loss of life (we spend tons of money extending the lives of dying people).
I was curious if there was an established set of principles that public health and other officials are trained to operate by. Do their models take loss of life, economic impact and other factors into consideration explicitly, taken together?There must be academic research in this regard. Here was one such model I found: https://budgetmodel.wharton.upenn.edu/issues/2020/5/1/coronavirus-reopening-simulator
As a data oriented person, I would vastly prefer to know what the effective loss function is that is being optimized for, even if I disagreed that it was the right one. I posited that if the public was clearer on what this was, that would also cut down on the politicization and make it easier to have everybody contribute to achieving the optimal outcome.