Question
This property appears to be overcome by enormously overparameterized models in an as-yet poorly understood phenomenon known as double descent. For 10 points each:
[10m] Name this property that arises because of a decomposition of mean-squared error as the sum of one quantity plus the square of another quantity. Overfitting and underfitting are both risks when ignoring this property.
ANSWER: bias-variance tradeoff [or bias-variance decomposition]
[10h] In a popular nonparametric method, this quantity should scale as “n to the negative one-fifth” so that both terms in the bias-variance decomposition are of the same order. This quantity divides “x minus xj (“x-sub-j”) in the input of a function used by the Nadaraya-Watson estimator.
ANSWER: kernel bandwidth (The method is kernel density estimation.)
[10e] A similar example is tuning the bin width for these diagrams to balance bias and variance. These diagrams are typically plotted as vertical frequency bars.
ANSWER: histograms
<Morrison, Other Science>
Summary
| 2024 ESPN @ Stanford | 03/09/2024 | Y | 2 | 10.00 | 100% | 0% | 0% |
| 2024 ESPN @ Brown | 04/06/2024 | N | 3 | 10.00 | 67% | 33% | 0% |
| 2024 ESPN @ Cambridge | 04/06/2024 | N | 2 | 10.00 | 100% | 0% | 0% |
| 2024 ESPN @ Chicago | 03/23/2024 | N | 6 | 11.67 | 83% | 33% | 0% |
| 2024 ESPN @ Columbia | 03/23/2024 | N | 7 | 14.29 | 86% | 57% | 0% |
| 2024 ESPN @ Duke | 03/23/2024 | N | 2 | 20.00 | 100% | 100% | 0% |
| 2024 ESPN @ Online | 06/01/2024 | N | 3 | 10.00 | 67% | 33% | 0% |
Data
| Berkeley B | Free Agents | 0 | 0 | 10 | 10 |
| Berkeley A | Stanford | 0 | 0 | 10 | 10 |