Specific Algorithmic Interpretability of Neural Networks: A Case Study on Textures

Excerpt

arXiv:2610.04413v1 Announce Type: cross Abstract: We develop a principled framework for constructing neural networks whose specific parameter realizations admit an explicit algorithmic interpretation. Existing algorithm-inspired architectures can explain the computational structure of a network, yet after standard training the learned parameters need not retain a clear relation to the motivating algorithm. We address this gap as follows. First, we model each data point as a sample of a class-dep