Neuro-Symbolic AI

I believe a neuro-symbolic approach is the only way to realize truthworthy reasoning, thus achieving true intelligence. It is probably the only hope to bridge the seemingly insurmountable gap between “correlation” and “causality”.

Here are some useful cases of the neural symbolic approach. And by now neuro-symbolic reasoning has become mainstream. Here is a podcast where I talked about AI reasoning.

  • Synthesizing high-quality, verifiably correct reasoning steps on solving math problems. (Li et al., 2024).
  • Improving the autoformalization, a fundamental math capability of a neural model, through a hybrid neural-symbolic solution. (Li et al., 2024).
  • Proving Olympiad inequalities through neuro-symbolic formal reasoning. (Li et al., 2025)
  • Spontaneous informal to formal reasoning. (Cao et al., 2025)
  • A neuro-symbolic approach to improve reasoning (Wang et al., 2026)

References

2026

  1. LoongRL: Reinforcement Learning for Advanced Reasoning over Long Contexts
    Siyuan Wang, and 6 more authors
    In International Conference on Learning Representations, ICLR, 2026

2025

  1. Proving Olympiad Inequalities by Synergizing LLMs and Symbolic Reasoning
    Zenan Li, and 8 more authors
    In International Conference on Learning Representations, ICLR, 2025
  2. Reviving DSP for Advanced Theorem Proving in the Era of Reasoning Models
    Chenrui Cao, and 6 more authors
    In Advances in Neural Information Processing Systems, NeurIPS, 2025

2024

  1. Neuro-Symbolic Data Generation for Math Reasoning
    Zenan Li, and 7 more authors
    In Advances in Neural Information Processing Systems, NeurIPS, 2024
  2. Autoformalize Mathematical Statements by Symbolic Equivalence and Semantic Consistency
    Zenan Li, and 6 more authors
    In Advances in Neural Information Processing Systems, NeurIPS, 2024