Surrogate Model

There is disagreement regarding the generalization ability of learned surrogate models: some views hold that learned models may struggle to generalize to unseen network topologies; others emphasize that generalization can be improved through physics-informed learning.

Lightweight learned surrogate models learn the simulator's input-output mapping through offline training and perform fast inference online, supporting online control at lower runtime cost.

4-bit quantization, gradient checkpointing, and small-scale proxy-model pre-experiments can bring down the GPU requirements of experiments during the queueing period, reducing the number of trials once the cards are actually obtained.

The paper verifies whether it can reproduce the simulator's output, not whether it proves the underlying biological mechanism is correct.

A latest paper proposes using a class-conditional diffusion model as a surrogate model to directly generate the final pattern of cell blood-vessel-like networks, compressing a single generation from hundreds of seconds to just over twenty seconds.

Using generative models as surrogates for simulators is becoming an active route in computational science.

A surrogate model capable of generating all possible solution branches at once can significantly accelerate uncertainty quantification and robustness assessment.

A surrogate model is a lightweight approximate model (such as a Gaussian process or neural network) trained with relatively few simulation samples, used to quickly predict the impact of design parameter changes on metrics such as power, area, and gain, thereby guiding the location of the next sampling point.

Fusionality uses AI surrogate models, employing neural networks to learn the input-output mapping of physics simulators, improving computational speed by several orders of magnitude while maintaining sufficient accuracy.