A convolutional layer maps 8 input channels to 16 output channels using 3×3 kernels. How many total trainable parameters (weights + biases) does this layer have?

Hot weather causes both higher ice cream sales and more drownings, creating a spurious correlation between the two if weather isn’t accounted for. This is an example of which causal DAG structure?

A convolutional layer uses far fewer parameters than a fully-connected layer mapping the same input to a same-sized output. This is mainly because of:

Which of the following IS actually guaranteed by the universal approximation theorem?

In a setting where treatment is not randomly assigned, the naive estimator (mean outcome of the treated minus mean outcome of the untreated) can be decomposed as the Average Treatment effect on the Treated (ATT) plus:

Which of the following is NOT actually true of ReLU (and therefore not a real reason it’s preferred over sigmoid/tanh for hidden layers)?

What is the formula for self-attention given queries Q, keys K, and values V?

In one iteration of Lloyd’s algorithm, what happens during the “update” step?

Why does convexity matter for gradient descent?

Required sample size per group in an A/B test scales as 1/delta^2, where delta is the minimum detectable effect. If you halve the minimum detectable effect you want to reliably detect (holding power and alpha fixed), the required sample size per group: