DisjunctiveNet: Neural Symbolic Learning via Differentiable Convexified Optimization Layers

ArXi:2605.30456v1 Announce Type: new Many learning tasks in science and engineering are characterized by sparse datasets, which limits the effectiveness of purely data-driven approaches. At the same time, these problems are often accompanied by rich domain knowledge derived from physical laws, operational requirements, and expert heuristics. Such knowledge is frequently expressed as rules involving logical propositions and linear inequalities.