202509282020 Status: #idea Tags: #reinforcement_learning #ai # Planning problems vs learning problems **Planning problems:** refers to problems in which a model of the environment is available and thus, there's no learning required. **Learning problems:** refers to problems in which learning from samples is required, usually because there isn't a model of the environment available or perhaps because it's impossible to create one. Learning problems can be further broken down into two categories: 1. **Non-interactive learning problems** - a type of learning problem in which there's no need for or possibility of interacting with an environment. In these types of problems, there's no interaction with an environment while learning, but there is learning from data. Examples include inverse RL, which is learning a reward function from expert interaction data, or imitation learning, which is directly learning a policy using supervised learning on top of expert data 2. **Interactive learning problems** - refers to a type of learning problem in which learning and interaction are interleaved. The interesting aspect of these problems is that the learner also controls the data-gathering process. --- # References [[Grokking Deep Reinforcement Learning]]