Dictionary of Applied Machine Learning
Updated on 2026-09-18
Typeset PDF version — the authoritative form of this entry
A reward is an observed or measured quantity that allows a machine learning (ML) method to estimate the loss incurred by the prediction or action of a hypothesis $\hypothesis(\featurevec)$. A reward replaces the label: it does not reveal the correct prediction, only how good the chosen one was, and it may be delayed or noisy. Reinforcement learning (RL) methods construct a loss from the rewards received and choose actions that maximize the cumulative reward. For example, a self-driving vehicle whose steering direction is $\hypothesis(\featurevec)$ can derive a reward from a collision sensor, assigning a low reward to a direction that moves the vehicle toward an obstacle.
A reward refers to some observed (or measured) quantity that allows estimation of the loss incurred by the prediction (or decision) of a hypothesis $\hypothesis(\featurevec)$. For example, in a machine learning (ML) application to self-driving vehicles, $\hypothesis(\featurevec)$ could represent the current steering direction of a vehicle. A reward could be constructed from the measurements of a collision sensor that indicate if the vehicle is moving toward an obstacle. A low reward is defined for the steering direction $\hypothesis(\featurevec)$ if the vehicle moves dangerously toward an obstacle.
See also: loss, multiarmed bandit, reinforcement learning.
@misc{dictml_reward,
author = {Jung, Alexander and Olioumtsevits, Konstantina and Schnoor, Ekkehard},
title = {reward},
howpublished = {Dictionary of Applied Machine Learning (course edition)},
year = {2026},
doi = {10.5281/zenodo.21569296},
note = {ISBN 978-952-64-3013-3, CC BY 4.0, retrieved 2026-09-19},
url = {https://dictionaryofml.org/terms/reward.html}
}