Implementing differential privacy in machine learning involves methods that integrate privacy protection directly into the learning process. Common techniques include:

- Output Perturbation: Adding noise to the output of a learning algorithm, effectively masking the influence of any single data point.

- Objective Perturbation: Altering the objective function of the learning algorithm by introducing a noise component, thus ensuring that the learning process itself preserves privacy.

- Gradient Perturbation: Particularly in algorithms like stochastic gradient descent, adding noise to the gradients used for learning to prevent any individual data point from having a significant influence on the model. These methods help in training machine learning models that not only respect the privacy of individual data points but also maintain good generalisation performance.