Flashcards · Reinforcement Learning · Free
Reinforcement Learning flashcards, generated for you.
Example Reinforcement Learning study cards to learn from right now — then generate a full set from your own notes (plus a practice quiz) and export to Quizlet or Anki. Free, no account needed.
Example Reinforcement Learning flashcards
What is the core idea of reinforcement learning?
An agent learns to make decisions by interacting with an environment, receiving rewards or penalties for actions, and adjusting behavior to maximize total reward over time.
What is the difference between reward and return in RL?
Reward is the immediate signal received after one action. Return is the cumulative sum of discounted rewards from a state forward (often written as G_t = R_t+1 + γR_t+2 + γ²R_t+3 + ...).
What does a policy do in reinforcement learning?
A policy π is a strategy that maps states to actions (or states to probability distributions over actions). It defines what the agent should do in each situation.
What is the Markov Property, and why does it matter in RL?
The Markov Property states that the future depends only on the current state, not on the history. This allows RL algorithms to make decisions based on the current state alone, making problems tractable.
What is the value function V(s), intuitively?
V(s) estimates how good it is to be in state s—it's the expected cumulative reward an agent will receive starting from that state, following the current policy.
What is the action-value function Q(s,a), and how does it differ from V(s)?
Q(s,a) estimates the expected cumulative reward for taking action a in state s, then following the policy. Unlike V(s), it conditions on a specific action, allowing the agent to compare which action is best.
What is the Bellman equation for V(s)?
V(s) = E[R_t+1 + γV(s')] — the value of a state equals the expected immediate reward plus the discounted value of the next state. This recursive relationship enables bootstrapping.
What is the exploration-exploitation tradeoff?
Exploration means trying new actions to discover better rewards; exploitation means choosing actions known to yield high rewards. The agent must balance both—pure exploitation risks missing better options, pure exploration wastes time.
How does Q-learning update estimates, and why is it called 'off-policy'?
Q-learning updates Q(s,a) using: Q(s,a) ← Q(s,a) + α[R + γ max Q(s',a') - Q(s,a)]. It's off-policy because it learns the optimal policy while following any exploratory policy (greedy w.r.t. max).
What is the key difference between value-based and policy-gradient methods?
Value-based methods (Q-learning, V-learning) learn value functions and derive policy implicitly (e.g., greedy). Policy-gradient methods directly parameterize and optimize the policy by gradient ascent on expected return.
Make your own Reinforcement Learning study set
Flashcards for related topics
Studying Reinforcement Learning to build with AI? MindloomHQ turns it into real skills — structured courses, agent projects, and certificates.
Explore MindloomHQ →