interpretabilitycritiquebearishNeural network-based operator learning architectures are opaque models that obscure the reasoning behind their predictionsMachine Learning27 Jul 2026http://arxiv.org/abs/2607.02203v1