Scholarlink

<p>The artificial intelligence system tends to be brittle, unstable in convergence, and opaque in ethical decision-making when exposed to dynamic or corrupted environments. The conventional neural networks are not endowed with intrinsic capabilities for self-healing or adaptive control, resulting in deterioration of performance and decision consistency. To address these issues, a biologically inspired and mathematically based artificial intelligence system that is capable of self-healing, foresight, and ethical decision-making is proposed in this paper. The proposed system is named <i>AI Brain Engine</i>, which is based on a biologically inspired and mathematically based framework for artificial intelligence system development. The system comprises a set of modular neural cells that are each provided with <i>LocalDNA</i> and <i>GlobalDNA</i> encoding capabilities. The system also comprises a reinforcement-guided <i>failure function</i> that automatically adjusts the learning rate based on error sensitivity. The system is also provided with a <i>Laplace-Demon-inspired module</i> that facilitates deterministic foresight capabilities for self-healing decision-making.The major focus is on evaluating the classification robustness in a degraded network environment using benchmark image dataset. Simulation tests on these image datasets have shown improved robustness and faster convergence compared to traditional deep learning networks like CNN and ResNet variants. The self-healing ability of the network is tested using network corruption protocols like weight noise, partial parameter reset, and pruning. In addition to these classification tests, further tests on the CMAPSS turbofan engine dataset are performed to evaluate the general applicability of the proposed framework for predictive maintenance. The results showed improvement in self-healing from network corruption by 9.3% in the resilience score and reduced volatility in network losses over training iterations. The results indicate that the integration of biological self-repair mechanisms into a deterministic computing paradigm results in a self-sustaining neural network that is capable of predicting degradation, optimizing response, and maintaining ethical integrity.</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

Bio-evolutionary neural networks with deterministic foresight and adaptive self-healing

  • 期刊论文

  • 发表日期：

    2026年6月27日

查看全文
  • R. Sachin,
  • K. Aasritha,
  • Vishnu Reddy,
  • Raghavendra Singh Jagawat,
  • Manava Daasardi,
  • S. Manimaran

摘要

The artificial intelligence system tends to be brittle, unstable in convergence, and opaque in ethical decision-making when exposed to dynamic or corrupted environments. The conventional neural networks are not endowed with intrinsic capabilities for self-healing or adaptive control, resulting in deterioration of performance and decision consistency. To address these issues, a biologically inspired and mathematically based artificial intelligence system that is capable of self-healing, foresight, and ethical decision-making is proposed in this paper. The proposed system is named AI Brain Engine, which is based on a biologically inspired and mathematically based framework for artificial intelligence system development. The system comprises a set of modular neural cells that are each provided with LocalDNA and GlobalDNA encoding capabilities. The system also comprises a reinforcement-guided failure function that automatically adjusts the learning rate based on error sensitivity. The system is also provided with a Laplace-Demon-inspired module that facilitates deterministic foresight capabilities for self-healing decision-making.The major focus is on evaluating the classification robustness in a degraded network environment using benchmark image dataset. Simulation tests on these image datasets have shown improved robustness and faster convergence compared to traditional deep learning networks like CNN and ResNet variants. The self-healing ability of the network is tested using network corruption protocols like weight noise, partial parameter reset, and pruning. In addition to these classification tests, further tests on the CMAPSS turbofan engine dataset are performed to evaluate the general applicability of the proposed framework for predictive maintenance. The results showed improvement in self-healing from network corruption by 9.3% in the resilience score and reduced volatility in network losses over training iterations. The results indicate that the integration of biological self-repair mechanisms into a deterministic computing paradigm results in a self-sustaining neural network that is capable of predicting degradation, optimizing response, and maintaining ethical integrity.