Differentiable Fuzzy Inference Layer: A Monotone, Compositional Ordinal Reasoning Head for Large Language Models
Read the original on arXiv Computation and Language →The paper introduces the Differentiable Fuzzy Inference Layer (DFIL), a dual‑path prediction head that pairs a standard classifier with a scalar‑bottlenecked branch using ordered membership functions. DFIL enforces monotonicity in the underlying quantity and enables compositional reasoning via t‑norm operations, addressing failures of standard classifier heads that treat ordinal categories as independent labels. The scalar branch also offers an interpretable interface for analyzing residual errors, and the authors demonstrate DFIL’s effectiveness on ordinal natural‑language tasks across various large language model families.
Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv Computation and Language.