Hyperparameter Trajectory Inference with Conditional Lagrangian Optimal Transport

Researchers introduce Hyperparameter Trajectory Inference (HTI), a method to predict how neural networks behave with different hyperparameter settings without expensive retraining. The approach uses conditional Lagrangian optimal transport to create surrogate models that approximate neural network outputs across various hyperparameter configurations.

• →HTI addresses the costly problem of retraining neural networks when user preferences change post-deployment.
• →The method uses conditional Lagrangian optimal transport to learn how neural network outputs change with different hyperparameters.
• →The approach incorporates manifold hypothesis and least-action principles to improve surrogate model feasibility.
• →Empirical results show superior performance in reconstructing neural network outputs compared to alternative methods.
• →The technique could reduce computational costs and time for adapting deployed AI models to new requirements.