Question 71
L2 Regularization (Ridge) penalizes model complexity by adding which term to the loss function?
Ridge regression adds an L2 squared norm penalty $\lambda \Sigma w_{i}^2, shrinking weight parameters towards zero$ to control variance.
L2 Regularization (Ridge) penalizes model complexity by adding which term to the loss function?
Ridge regression adds an L2 squared norm penalty $\lambda \Sigma w_{i}^2, shrinking weight parameters towards zero$ to control variance.
Which machine learning algorithm finds a hyper-plane that maximizes the margin between different classes of data points?
Support Vector Machines construct optimal decision boundary hyperplanes by maximizing the geometric margin to the closest training support vectors.
The activation function Leaky ReLU improves upon standard ReLU by:
Leaky ReLU defines f(x) = max($\alpha x, x$) where $\alpha \approx 0.01$ for x < 0, ensuring backpropagation gradients do not saturate to zero.
Which activation function is standard in the output layer of multi-class classification neural networks to produce normalized probability distributions?
Softmax $\exp(z_{i}) / \Sigma \exp(z_{j}) normalizes a$ vector of K real scores into a probability distribution summing to 1.0.
In training deep networks, Vanishing Gradient Problem primarily affects early network layers when using which activation functions?
Sigmoid and Tanh derivatives peak at $\le 0.25; multiplying$ these small fractions across many layers causes gradients to vanish exponentially.