A data scientist is configuring a SageMaker automatic model tuning job for an XGBoost model. The 'learning_rate' hyperparameter should be explored across values that span several orders of magnitude (from 0.001 to 0.5), and early experiments show that most performance improvements occur at the smaller end of the range. The team wants the tuner to sample the search space efficiently so it does not waste trials concentrating on large values. How should the learning_rate parameter be defined in the tuning job?