Skip to content
ml-recall

#jensen-inequality

3 questions from Information Theory, Decision Trees & Ensembles, Probabilistic & Graphical Models

Timed test

3 questions · 3 min · score and explanations at the end

Take a test
Q 1 / 3Score 0 / 0

Question 1: Short answer

mediumAI-checkedInformation Theory

Complete the proof that DKL(p∥q)≥0D_{\mathrm{KL}}(p \| q) \ge 0 (assume pp and qq are positive on the same outcomes). Write the divergence as Ex∼p[−log⁡q(x)p(x)]E_{x \sim p}\left[-\log \frac{q(x)}{p(x)}\right]. Because −log⁡-\log is convex, Jensen's inequality moves the expectation inside the logarithm: DKL(p∥q)≥−log⁡Ex∼p[q(x)p(x)]D_{\mathrm{KL}}(p \| q) \ge -\log E_{x \sim p}\left[\frac{q(x)}{p(x)}\right]. The expectation inside the logarithm equals ∑xq(x)\sum_x q(x), which is . So the right-hand side equals , and the divergence can never be negative.

Fill in every blank (hints in the boxes). Enter moves to the next blank.