📄 中文摘要
研究提出了一个统一框架，以更好地理解各种生物启发模型的结构和功能差异。研究表明，液体电容扩展模型即使在密集的全连接递归神经网络（RNN）策略中也能表现出可解释性。进一步证明了引入化学突触能够提高可解释性，并且将化学突触与突触激活相结合，能够产生最准确且可解释的RNN模型。为了评估这些RNN策略的准确性和可解释性，研究选择了具有挑战性的车道保持控制任务，并通过多个指标进行性能评估，包括转向加权验证损失、驾驶过程中的神经活动以及绝对相关性等。
突触激活与双液体动力学在可解释生物启发模型中的应用
📄 English Summary
Synaptic Activation and Dual Liquid Dynamics for Interpretable Bio-Inspired Models
This research presents a unified framework to better understand the structural and functional differences among various bio-inspired models. It demonstrates that liquid-capacitance-extended models exhibit interpretable behavior even in dense, all-to-all recurrent neural network (RNN) policies. Additionally, the incorporation of chemical synapses is shown to enhance interpretability, while the combination of chemical synapses with synaptic activation yields the most accurate and interpretable RNN models. To evaluate the accuracy and interpretability of these RNN policies, the challenging lane-keeping control task is considered, and performance is assessed across multiple metrics, including turn-weighted validation loss, neural activity during driving, and absolute correlation.