Reinforcement learning
A learning method in which a system develops an action strategy from actions and rewards in an environment.
How does a system learn through rewards?
In reinforcement learning, or RL, a system takes actions in an environment and receives associated rewards. It learns an action strategy, also called a policy, from this process. The objective is to maximise expected reward over the course being considered. Reward design is therefore crucial: what is rewarded computationally must fit the actual purpose of the task. A trained strategy and its later use need separate planning.
In a fictional simulation, a system is to learn sequences of service visits. A reward based only on short travel distances would not account for agreed time windows by itself. The team therefore defines permitted actions, mandatory planning constraints and suitable evaluation scenarios. It also examines situations where short routes conflict with completing all assigned visits on time. A high simulation score alone does not establish a suitable operational plan. Before practical use, the actual conditions and the handling of deviations are assessed separately.
Sources
Related terms
- Machine learningModels learn patterns from data and apply them to new inputs.
- AI inferenceApplying an already trained model to inputs to obtain a result.
- LLM evaluationLLM evaluation tests model outputs against defined tasks and criteria. One good example does not establish overall quality.
- Supervised learningA learning method that uses inputs together with their corresponding target values or classes as training examples.
- Unsupervised learningA learning method that looks for patterns in data without predefined target answers for the particular training task.
- Re-rankingRe-ranking reassesses search candidates already retrieved. Relevance to the query and the limits of the initial selection both matter.
Your project with ASES
Describe your project to us. We will get back to you with an initial assessment.