Which technique uses artificial neural networks to estimate values instead of storing them in a lookup table?

Which technique uses artificial neural networks to estimate values instead of storing them in a lookup table?

• A. Q-Learning
• B. Monte Carlo method
• C. Deep Reinforcement Learning
• D. Temporal Difference (TD) method

Answer: (C)

Estimating values with neural networks is the hallmark of deep reinforcement learning. In traditional Q-learning, you store a table of Q-values for every state-action pair, which becomes impractical as the state space grows or becomes continuous. Deep reinforcement learning replaces that table with a neural network that approximates the value function (or the policy). This lets the agent handle high-dimensional or continuous inputs and generalize to unseen states by learning patterns from data rather than memorizing every possible state-action value. A classic example is a neural network that takes the current state and outputs Q-values for each action, trained to predict future rewards from experiences. The other techniques don’t inherently rely on neural networks to estimate values. Q-learning in its standard form uses a table; Monte Carlo methods estimate returns from sampled episodes without a built-in neural network; and temporal difference methods are a broader idea for bootstrapping value estimates, which can be used with or without function approximation, not specifically with neural nets.

Estimating values with neural networks is the hallmark of deep reinforcement learning. In traditional Q-learning, you store a table of Q-values for every state-action pair, which becomes impractical as the state space grows or becomes continuous. Deep reinforcement learning replaces that table with a neural network that approximates the value function (or the policy). This lets the agent handle high-dimensional or continuous inputs and generalize to unseen states by learning patterns from data rather than memorizing every possible state-action value. A classic example is a neural network that takes the current state and outputs Q-values for each action, trained to predict future rewards from experiences.

The other techniques don’t inherently rely on neural networks to estimate values. Q-learning in its standard form uses a table; Monte Carlo methods estimate returns from sampled episodes without a built-in neural network; and temporal difference methods are a broader idea for bootstrapping value estimates, which can be used with or without function approximation, not specifically with neural nets.