Scholarlink

<p>Safety–critical systems must reconcile predictive accuracy with interpretability under shifting data distributions. We present a hybrid neuro-symbolic (HNS) framework coupling a residual neural encoder with a differentiable fuzzy constraint layer and an online Bayesian rule-revision mechanism, plus an information-theoretic explainability index. On four benchmarks the model attains 95.1% accuracy, 91.2% F1, explainability E = 0.91, and sub-8&#xa0;ms inference, surpassing neural, symbolic, LIME, and DeepProbLog baselines while remaining robust to distributional shift.</p>

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

Hybrid Neuro-Symbolic Models for Transparent and Adaptive Decision-Making in Dynamic Real-World Environments

  • 期刊论文

  • 发表日期：

    2026年7月17日

查看全文
  • Amit Welekar,
  • Abhrendu Bhattacharya,
  • Vishal Tiwari,
  • Swati Shamkuwar,
  • Mohammad Tahir,
  • Raju Pawar,
  • Swati Tiwari

摘要

Safety–critical systems must reconcile predictive accuracy with interpretability under shifting data distributions. We present a hybrid neuro-symbolic (HNS) framework coupling a residual neural encoder with a differentiable fuzzy constraint layer and an online Bayesian rule-revision mechanism, plus an information-theoretic explainability index. On four benchmarks the model attains 95.1% accuracy, 91.2% F1, explainability E = 0.91, and sub-8 ms inference, surpassing neural, symbolic, LIME, and DeepProbLog baselines while remaining robust to distributional shift.