Which approach uses a parameter to determine whether to explore or exploit at a given time?

• A. Policy-based approach
• B. Epsilon Greedy Strategy
• C. Value-based approach
• D. Deep Reinforcement Learning

Answer: (B)

Balancing exploration and exploitation is the idea here. The approach that uses a parameter to decide whether to explore or exploit at a given moment is the epsilon-greedy strategy. It keeps estimates of action values and, at each decision, picks a random action with probability epsilon, or the best-known action with probability 1 minus epsilon. The epsilon parameter directly controls how often the agent explores: a larger epsilon means more exploration, a smaller epsilon means more exploitation (and it can be decayed over time as learning progresses).

Other categories describe broader learning methods and may incorporate exploration in various ways, but they aren’t defined by a single explicit explore-or-exploit switch governed by a parameter in the way epsilon-greedy is.