Reward shaping
Reward shaping adds learning guidance. Poorly designed rewards can teach an agent to optimize a proxy while failing the intended task.
Definition
Reward shaping adds learning guidance. Poorly designed rewards can teach an agent to optimize a proxy while failing the intended task.
Intuition
Rewarding long answers may produce verbosity instead of accuracy. A reward must reflect what you actually want to improve.
Worked example
A maze agent rewarded whenever it approaches a goal may exploit repeated movements unless the shaping rule is designed carefully.
The math
Potential-based shaping uses , with assumptions and terminal handling needed for policy invariance.
In practice
Evaluate real task outcomes separately from the optimized reward and keep explicit failure examples in the test set.
Go deeper
Sources
More in Reinforcement learning
Assembled from the ReLU.chat curated knowledge base. These explanations are concise on purpose; check the sources for anything important.