BIRDNet: Mining and Encoding Boolean Implication Knowledge Graphs as Interpretable Deep Neural Networks

BIRDNet is an interpretable neurosymbolic neural network that mines Boolean implication relationships directly from tabular data and encodes them as sparse network connectivity. On transcriptomic and proteomic benchmarks, it achieves near-dense-model performance with up to 96× fewer active parameters while yielding biologically meaningful, human-readable rules.

Recommended citation: Dash, T. (2026). BIRDNet: Mining and Encoding Boolean Implication Knowledge Graphs as Interpretable Deep Neural Networks. arXiv preprint arXiv:2605.28739.
Download Paper