Prerequisites for DPO vs GRPO vs RL for Reasoning

Direct prerequisites (2)

1. RLHF and Alignmentlayer 4, tier 2
2. Policy Gradient Theoremlayer 3, tier 1

Reachable through the chain (14)

These topics are not directly cited as prerequisites but are reached transitively by following the chain upward. Working through the direct prerequisites pulls these in.

• Markov Decision Processeslayer 2, tier 1
• Convex Optimization Basicslayer 1, tier 1
• Differentiation in Rnlayer 0A, tier 1
• Sets, Functions, and Relationslayer 0A, tier 1
• Basic Logic and Proof Techniqueslayer 0A, tier 2
• Vectors, Matrices, and Linear Mapslayer 0A, tier 1
• Continuity in Rⁿlayer 0A, tier 1
• Metric Spaces, Convergence, and Completenesslayer 0A, tier 1
• Matrix Operations and Propertieslayer 0A, tier 1
• Concentration Inequalitieslayer 1, tier 1
• Common Probability Distributionslayer 0A, tier 1
• Expectation, Variance, Covariance, and Momentslayer 0A, tier 1
• Random Variableslayer 0A, tier 1
• Kolmogorov Probability Axiomslayer 0A, tier 1