Nadaraya Watson

The Nadaraya-Watson estimator, a non-parametric regression method, is experiencing renewed interest due to its computational efficiency and adaptability in various machine learning contexts. Current research focuses on improving its robustness against adversarial attacks and integrating it into more complex models, such as random forests and neural networks, often leveraging attention mechanisms to enhance performance. This renewed focus stems from its ability to provide explainable and calibrated predictions, particularly valuable in applications requiring safety guarantees (e.g., robotics) or interpretability (e.g., medical diagnosis), while offering computational advantages over more complex alternatives.