ML-Quiz 1

Ridge and Lasso regression are simple techniques to ________ the complexity of the model and prevent over-fitting which may result from simple linear regression

• Increase

• Decrease

• Eliminate

• None of the given

Increase in the hyperparameter in Ridge regression results in ?

• Larger bias, larger variance

• Larger bias, smaller variance

• Smaller bias, larger variance

• Smaller bias, smaller variance

Suppose that we are trying to fit a linear and 10th degree polynomial to data coming from a cubic function, corrupted by standard Gaussian noise. Let M1 and M2 denote the models corresponding to the linear and 10 degree polynomial. Then,

• Bias(M1) ≤ Bias(M2), Variance(M1) ≤ Variance(M2)

• Bias(M1) ≤ Bias(M2), Variance(M1) ≥ Variance(M2).

• Bias(M1) ≥ Bias(M2), Variance(M1) ≤ Variance(M2).

• Bias(M1) ≥ Bias(M2), Variance(M1) ≥ Variance(M2).

What causes noise?

• Error with the model

• Error with the data

Which of the formula characterizes expected loss interms of bias variance decomposition?

• bias * variance + noise

• bias² + variance + noise

• bias² * variance + noise

• bias + variance² + noise