High Quality Embeddings for Horn Logic Reasoning

Yifan Zhang, Yasir White, Dean Clark, Joseph Sanchez, Jevon Lipsey, Ashely Hirst, Jeff Heflin

arXiv:2605.20467 · 2026-05-22 공개 · arXiv · PDF

logical-reasoning knowledge-base triplet-loss embedding-learning horn-logic reasoning-ai neural-embeddings training-strategies

Abstract

Neural networks can be trained to rank the choices made by logical reasoners, resulting in more efficient searches for answers. A key step in this process is creating useful embeddings, i.e., numeric representations of logical statements. This paper introduces and evaluates several approaches to creating embeddings that result in better downstream results. We train embeddings using triplet loss, which requires examples consisting of an anchor, a positive example, and a negative example. We introduce three ideas: generating anchors that are more likely to have repeated terms, generating positive and negative examples in a way that ensures a good balance between easy, medium, and hard examples, and periodically emphasizing the hardest examples during training. We conduct several experiments to evaluate this approach, including a comparison of different embeddings across different knowledge bases, in an attempt to identify what characteristics make an embedding well-suited to a particular reasoning task.

한국어 요약

한 줄 요약

**[Horn Logic / Neural Embedding]** Triplet loss 기반 logical statement 임베딩에 anchor 반복 항·easy/medium/hard 균형 예제·hard example 강조 3 아이디어 도입, 다양 KB·reasoning task에서 임베딩 적합성 특성 식별.

핵심 기여도

핵심 아이디어

Logical statement 임베딩 품질은 단순 triplet loss로는 부족하며, anchor의 구조적 특성(반복 항)·예제 난이도 분포 균형·hard example 강조의 3 학습 신호 결합이 downstream reasoning 성능을 결정한다.

기술적 접근법

주요 결과

의의 및 한계

**의의**: Horn logic reasoning의 신경 가속에서 임베딩 품질을 직접 다룬 시도, triplet loss의 단순 적용 한계를 3 학습 신호로 보완, KB·task별 특성화 가능성 제시. **한계**: 정확한 정량 향상 abstract 미명시, Horn logic·특정 reasoner에 한정으로 first-order·description logic 일반화 미검증, 임베딩 차원·KB 규모별 scalability는 후속.

실용적 활용