Trained quantum neural networks have shown to mimic Gaussian processes in infinite width scenarios, as outlined in a recent study. The paper provides insights on:

• The behavior of untrained networks with random parameters as Gaussian processes
• How trained networks using gradient descent can perfectly fit training data
• The sufficiency of polynomial measurements for these networks to function effectively

This research indicates that trained QNNs can offer accurate predictive models with practical training requirements. From my perspective, the Gaussian process view aids comprehensively in understanding the underlying mechanisms of QNNs, highlighting their potential scalability and effectiveness.