Toward Verifiable and Scalable In‐context Fine‐tuning in Neurosymbolic AI

View at publisher Summarise with VIRAT
Authors: Neel Pratik Bhatt, Alvaro Velasquez, Zhangyang Wang, Ufuk Topcu
Cited by: 0
References: 39

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Scaling laws have taken center stage in recent artificial intelligence (AI) research owing to the increasing availability of computer and training data. However, this approach may soon reach theoretical and practical limits. Neurosymbolic AI offers a more efficient path forward by utilizing structured symbolic representations, prior knowledge, and logic in conjunction with the power of neural computation enabling scalability and verifiability. Notably, planning is a field where purely neural models struggle to maintain logical consistency and compliance with task specifications. By applying neurosymbolic methods that perform output reasoning and input prompting, we can bridge the gap between neural learning and logical reasoning offering verifiable and scalable planning solutions that neither approach can achieve independently. We present two verifiable and scalable representative neurosymbolic frameworks that perform output reasoning and subsequent model fine-tuning. In addition, we provide insights into a hybrid reasoning and prompting framework followed by future avenues for research that highlight the importance of developing novel neurosymbolic AI frameworks in delivering verifiable and scalable models.

Fields of Study & Subjects

AI-based Problem Solving and Planning Logic Reasoning and Knowledge Artificial Intelligence Applications

Publication details

Journal / source
Wiley eBooks
Pages
223–266
Publication date
20 March 2026
Language
EN
Document type
Book chapter