Kernel Based Surrogate

Kernel-based surrogate models aim to efficiently approximate complex functions, particularly those arising from computationally expensive methods like deep neural networks or quantum algorithms. Current research focuses on developing improved surrogate architectures, such as those employing kernel sum of squares or conjugate kernels, and optimizing their performance through global optimization techniques and sketching methods to reduce computational cost. These surrogates offer significant advantages by accelerating training and inference, enabling the application of powerful but computationally demanding models to larger datasets and more complex problems across diverse fields, including dynamical systems modeling and machine learning.