Is There Always a Tradeoff Between Bias and Variance?

The mean squared error (<a href="http://bit.ly/quaesita_babymse" rel="noopener ugc nofollow" target="_blank">MSE</a>) is the most popular (and vanilla) choice for a model’s <a href="http://bit.ly/quaesita_emperorm" rel="noopener ugc nofollow" target="_blank">loss function</a> and it tends to be the first one you’re taught. You’ll likely take a whole bunch of <a href="http://bit.ly/quaesita_statistics" rel="noopener ugc nofollow" target="_blank">stats</a> classes before it occurs to anyone to tell you that you’re welcome to minimize other loss functions if you like. (But let’s be real: <a href="http://bit.ly/quaesita_msefav" rel="noopener ugc nofollow" target="_blank">parabolae are super easy to optimize</a>. Remember d/dx x²? 2x. That convenience is enough to keep most of you loyal to the MSE.) Once you learn about the MSE, it’s usually mere <a href="http://bit.ly/quaesita_lemur" rel="noopener ugc nofollow" target="_blank">moments</a> until someone mentions the bias and variance formula: <a href="https://towardsdatascience.com/is-there-always-a-tradeoff-between-bias-and-variance-5ca44398a552">Website</a>