Research Connects Binarized Neural Networks to Sugeno Integral Framework

A research paper establishes a formal relationship between binarized neural networks and Sugeno integrals. This mathematical framework allows the representation of input importance and interactions through equivalent if-then rules. For hidden neurons during inference, activation threshold tests can be expressed as Sugeno integrals on binary inputs. Each neuron's decision receives an explicit set-function representation alongside a rule-based interpretation. The final layer score also obtains a Sugeno-integral formulation. The approach can be adapted to handle more complex input interactions and extended beyond binary contexts. The work was published on arXiv with identifier 2604.17967.