Page 120 - 《软件学报》2026年第2期

丁嵘 等: 基于领域知识图谱的框架间 AI 源码自动迁移 599

表 14 模型精度测试结果 (%)

网络名称 PyTorch PaddlePaddle MindSpore
ResNet50 94.28 94.36 94.65
MobileNetV2 91.29 91.67 90.85
DenseNet121 94.11 93.61 93.76
ViT 53.13 54.28 56.36
ShuffleNet V2 89.17 88.66 88.72
HarDNet 87.51 86.75 87.51
EfficientNet 90.26 89.54 89.66
Gather-excite 94.49 94.51 94.53
U-Net 71.59 72.63 74.01
PSPNet_ResNet50 62.17 63.28 64.38

6 总结与展望

本文提出了一种基于领域知识图谱和抽象语法树的不同深度学习框架间 AI 源码迁移方法. 通过建立领域知
识图谱来存储不同深度学习框架间算子及其参数的差异信息, 并采用抽象语法树作为代码迁移中间件提高了转换
的可靠性与效率. 实验结果表明, 我们的方法在不同深度学习框架间的 AI 源码自动迁移上具有较好的转换效果,
模型功能性测试证明了代码迁移的正确性, 模型精度测试进一步证明了迁移后的模型的可靠性.
在接下来的工作中, 我们考虑完善迁移代码的完整性, 包括模型的训练代码、测试代码以及数据集读取代码
等. 上述代码存在于模型构建中除模型设计外的其余步骤中, 如数据处理、训练设置、应用部署和模型评估, 其中
包含若干多对多映射关系的算子. 我们计划采用基于自定义算子的组合替代实现来进行多对多算子的迁移, 并将
该自定义算子覆盖到源抽象语法树中来达到改写抽象语法树的目的. 最后, 我们计划进一步完善深度学习框架的
数量, 增加其他框架下模型的迁移方案, 实现更多框架之间的互联互通, 以凝聚深度学习框架间的群智力量.

References
[1] Paszke A, Gross S, Massa F, Lerer A, Bradbury J, Chanan G, Killeen T, Lin ZM, Gimelshein N, Antiga L, Desmaison A, Köpf A, Yang
E, DeVito Z, Raison M, Tejani A, Chilamkurthy S, Steiner B, Fang L, Bai JJ, Chintala S. PyTorch: An imperative style, high-
performance deep learning library. In: Proc. of the 33rd Int’l Conf. on Neural Information Processing Systems. Vancouver: Curran
Associates Inc., 2019. 8026–8037.
[2] Ma YJ, Yu DH, Wu T, Wang HF. PaddlePaddle: An open-source deep learning platform from industrial practice. Frontiers of Data and
Computing, 2019, 1(1): 105–115 (in Chinese with English abstract). [doi: 10.11871/jfdc.issn.2096.742X.2019.01.011]
[3] Shridhar A, Tomson P, Innes M. Interoperating deep learning models with ONNX. Proc. of the JuliaCon Conf., 2020, 1(1): 59. [doi: 10.
21105/jcon.00059]
[4] Liu Y, Chen C, Zhang R, Qin TT, Ji X, Lin HX, Yang M. Enhancing the interoperability between deep learning frameworks by model
conversion. In: Proc. of the 28th ACM Joint Meeting on European Software Engineering Conf. and Symp. on the Foundations of
Software Engineering. ACM, 2020. 1320–1330. [doi: 10.1145/3368089.3417051]
[5] Jia YQ, Shelhamer E, Donahue J, Karayev S, Long J, Girshick R, Guadarrama S, Darrell T. Caffe: Convolutional architecture for fast
feature embedding. In: Proc. of the 22nd ACM Int’l Conf. on Multimedia. Orlando: ACM, 2014. 675–678. [doi: 10.1145/2647868.
2654889]
[6] Abadi M, Barham P, Chen JM, Chen ZF, Davis A, Dean J, Devin M, Ghemawat S, Irving G, Isard M, Kudlur M, Levenberg J, Monga R,
Moore S, Murray DG, Steiner B, Tucker P, Vasudevan V, Warden P, Wicke M, Yu Y, Zheng XQ. TensorFlow: A system for large-scale
machine learning. In: Proc. of the 12th USENIX Conf. on Operating Systems Design and Implementation. Savannah: USENIX
Association, 2016. 265–283.
[7] Seide F, Agarwal A. CNTK: Microsoft’s open-source deep-learning toolkit. In: Proc. of the 22nd ACM SIGKDD Int’l Conf. on
Knowledge Discovery and Data Mining. San Francisco: ACM, 2016. 2135. [doi: 10.1145/2939672.2945397]
[8] Ren S, Guo D, Lu S. CodeBLEU: A method for automatic evaluation of code synthesis. arXiv:2009.10297, 2020.
[9] Roziere B, Lachaux MA, Chanussot L, Lample G. Unsupervised translation of programming languages. In: Proc. of the 34th Int’l Conf.