标题： 基于$MoS_2$基于闪存的模拟CAM的可信树型机器学习，具有固有的软边界 显示英文标题

标题： Trustworthy Tree-based Machine Learning by $MoS_2$ Flash-based Analog CAM with Inherent Soft Boundaries

摘要： 人工智能的快速发展引发了对其可信度的担忧，尤其是在可解释性和鲁棒性方面。 基于树的模型如随机森林和XGBoost在表格数据的可解释性和准确性方面表现出色，但由于数据局部性差和数据依赖性高，扩展它们仍然计算成本高昂。 以前尝试使用模拟内容可寻址存储器（CAM）加速这些模型的努力遇到了困难，因为难以实现的尖锐决策边界对设备变化非常敏感，这导致硬件性能不佳且容易受到对抗攻击。 本工作提出了一种新颖的软硬件协同设计方法，使用$MoS_2$ 基于闪存的模拟CAM，具有固有的软边界，使软树模型能够高效推理。 我们在$MoS_2$模拟CAM阵列上的软树模型推理实验表明，这种方法在保持最先进的准确性的同时，对设备变化和对抗攻击具有极强的鲁棒性。 具体来说，我们制造的模拟CAM阵列在威斯康星诊断乳腺癌（WDBC）数据库上实现了$96\%$的准确率，同时保持了决策可解释性。 我们的实验校准模型在 MNIST 数据集上，在$10\%$设备阈值变化下仅出现了$0.6\%$的准确率下降，相比之下，传统决策树出现了$45.3\%$的下降。 这项工作为增强人工智能可信度和效率的专业化硬件铺平了道路。 显示英文摘要

摘要： The rapid advancement of artificial intelligence has raised concerns regarding its trustworthiness, especially in terms of interpretability and robustness. Tree-based models like Random Forest and XGBoost excel in interpretability and accuracy for tabular data, but scaling them remains computationally expensive due to poor data locality and high data dependence. Previous efforts to accelerate these models with analog content addressable memory (CAM) have struggled, due to the fact that the difficult-to-implement sharp decision boundaries are highly susceptible to device variations, which leads to poor hardware performance and vulnerability to adversarial attacks. This work presents a novel hardware-software co-design approach using $MoS_2$ Flash-based analog CAM with inherent soft boundaries, enabling efficient inference with soft tree-based models. Our soft tree model inference experiments on $MoS_2$ analog CAM arrays show this method achieves exceptional robustness against device variation and adversarial attacks while achieving state-of-the-art accuracy. Specifically, our fabricated analog CAM arrays achieve $96\%$ accuracy on Wisconsin Diagnostic Breast Cancer (WDBC) database, while maintaining decision explainability. Our experimentally calibrated model validated only a $0.6\%$ accuracy drop on the MNIST dataset under $10\%$ device threshold variation, compared to a $45.3\%$ drop for traditional decision trees. This work paves the way for specialized hardware that enhances AI's trustworthiness and efficiency.