Hyperparameter Learning for Conditional Kernel Mean Embeddings with Rademacher Complexity Bounds
This work addresses a specific bottleneck in nonparametric probabilistic inference for researchers and practitioners using conditional kernel mean embeddings, offering a more efficient and scalable hyperparameter tuning method.
The paper tackles the problem of hyperparameter tuning for conditional kernel mean embeddings, which is currently reliant on expensive cross-validation or heuristics, by proposing a learning framework based on Rademacher complexity bounds to prevent overfitting. The result shows that this framework outperforms competing methods and can be extended to incorporate deep neural network weights for improved generalization.
Conditional kernel mean embeddings are nonparametric models that encode conditional expectations in a reproducing kernel Hilbert space. While they provide a flexible and powerful framework for probabilistic inference, their performance is highly dependent on the choice of kernel and regularization hyperparameters. Nevertheless, current hyperparameter tuning methods predominantly rely on expensive cross validation or heuristics that is not optimized for the inference task. For conditional kernel mean embeddings with categorical targets and arbitrary inputs, we propose a hyperparameter learning framework based on Rademacher complexity bounds to prevent overfitting by balancing data fit against model complexity. Our approach only requires batch updates, allowing scalable kernel hyperparameter tuning without invoking kernel approximations. Experiments demonstrate that our learning framework outperforms competing methods, and can be further extended to incorporate and learn deep neural network weights to improve generalization.