Scholarlink

Neural-symbolic (NeSy) AI has gained a lot of popularity by enhancing learning models with explicit reasoning capabilities. Both new systems and new benchmarks are constantly introduced and used to evaluate learning and reasoning skills. The large variety of systems and benchmarks, however, makes it difficult to establish a fair comparison among the various frameworks, let alone a unifying set of benchmarking criteria. This paper analyzes the state-of-the-art in benchmarking NeSy systems, studies its limitations, and proposes ways to overcome them. We categorize popular neural-symbolic frameworks into three groups: model-theoretic, proof-theoretic fuzzy, and proof-theoretic probabilistic systems. We show how these three categories have distinct strengths and weaknesses, and how this is reflected in the type of tasks and benchmarks to which they are applied.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Benchmarking in Neuro-Symbolic AI

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Robin Manhaeve,
  • Francesco Giannini,
  • Mehdi Ali,
  • Damiano Azzolini,
  • Alice Bizzarri,
  • Andrea Borghesi,
  • Samuele Bortolotti,
  • Luc De Raedt,
  • Devendra Dhami,
  • Michelangelo Diligenti,
  • Sebastijan Dumančić,
  • Boi Faltings,
  • Elisabetta Gentili,
  • Alfonso Gerevini,
  • Marco Gori,
  • Tias Guns,
  • Martin Homola,
  • Kristian Kersting,
  • Jens Lehmann,
  • Michele Lombardi,
  • Luca Lorello,
  • Emanuele Marconato,
  • Stefano Melacci,
  • Andrea Passerini,
  • Debjit Paul,
  • Fabrizio Riguzzi,
  • Stefano Teso,
  • Neil Yorke-Smith,
  • Marco Lippi

摘要

Neural-symbolic (NeSy) AI has gained a lot of popularity by enhancing learning models with explicit reasoning capabilities. Both new systems and new benchmarks are constantly introduced and used to evaluate learning and reasoning skills. The large variety of systems and benchmarks, however, makes it difficult to establish a fair comparison among the various frameworks, let alone a unifying set of benchmarking criteria. This paper analyzes the state-of-the-art in benchmarking NeSy systems, studies its limitations, and proposes ways to overcome them. We categorize popular neural-symbolic frameworks into three groups: model-theoretic, proof-theoretic fuzzy, and proof-theoretic probabilistic systems. We show how these three categories have distinct strengths and weaknesses, and how this is reflected in the type of tasks and benchmarks to which they are applied.