结合改进高斯核的图半监督转导小样本学习

doi:10.3778/j.issn.1002-8331.2207-0441

摘要/Abstract

摘要： 近年来，深度学习在机器学习领域取得巨大的研究进展，广泛应用于各个行业。但其需要大量地标注数据训练模型，资源成本耗费较大。因此，小样本学习逐渐成为机器学习的研究热点之一，并可结合半监督学习解决小样本学习标注数据少的问题。为了提高小样本学习模型准确率，针对半监督转导传播网络模型中高斯核在无限远处的衰减几乎为零的问题，提出适用于半监督转导网络模型的改进高斯核函数。通过加入位移参数和修正参数，使其在高维特征空间中能在测试点附近具有较快的衰减速度且在无限远处仍能保持适度的衰减，提高了小样本学习模型效果。在监督和半监督环境下进行实验对比，实验结果表明该算法在一定程度上提高模型精度，且具有实用价值。

关键词: 小样本学习, 元学习, 改进高斯核, 图半监督学习, 标签传播

Abstract: Few-shot learning is one of the most significant research in machine learning. In recent years, deep learning has made great progress in machine learning and is widely used in various industries. However, it requires a large amount of annotated data to train the model, which costs a lot of resources. Therefore, few-shot learning has gradually become one of the research hotspots of machine learning. To improve the accuracy of the few-shot learning model, in this paper, a novel graph semi-supervised transduction propagation network model with an improved Gaussian kernel function（IG-semiTPN） is presented to solve the problem that the attenuation rate of the Gaussian kernel is almost zero in infinite. The parameters of displacement and correction are introduced to the model to make it have a fast attenuation rate near the test point and maintain a moderate attenuation in the infinite in the high-dimensional feature space, to improve the effectiveness of the few-shot learning model. Experimental results show that the proposed approach can improve the accuracy to a certain extent and has practical value.

Key words: few-shot learning, meta learning, improved Gauss kernel, graph semi-supervision learning, label propagation

潘雪玲, 李国和, 于秋月, 郭凯, 李铮. 结合改进高斯核的图半监督转导小样本学习[J]. 计算机工程与应用, 2023, 59(17): 328-333.

PAN Xueling, LI Guohe, YU Qiuyue, GUO Kai, LI Zheng. Novel Graph Semi-Supervised Transduction Approach with Improved Gauss Kernel for Few-Shot Learning[J]. Computer Engineering and Applications, 2023, 59(17): 328-333.

参考文献

[1] CHEN Z T，FU Y W，ZHANG Y D，et al.Semantic feature augmentation in few-shot learning[C]//European Conference on Computer，2018.
[2] LI H Y，DONG W M，MEI X，et al.LGM-Net：learning to generate matching networks for few-shot learning[C]//International Conference on Machine Learning，2019：3825-3834.
[3] SNELL J，SWERSKY K，ZEMEL R.Prototypical networks for few shot learning[C]//Neural Information Processing System，2017：4077-4087.
[4] LIU Y，LEE J，PARK M，et al.Learning to propagate labels：transductive propagation network for few-shot learning[C]//International Conference on Learning Representations，2019.
[5] 赵曙光，杨万海.基于函数级FPGA原型的硬件内部进化[J].计算机学报，2002，25（6）：666-669.
ZHAO S G，YANG W H. Intrinsic hardware evolution based on a prototype of function level FPGA[J].Chinese Journal of Computers，2002，25（6）：666-669.
[6] REMAKI L，CHERIET M.KCS-new kernel family with compact support in scale space：formulation and impact[J].IEEE Transactions on Image Processing，2000，9（6）：970-981.
[7] VINYALS O，BLUNDELL C，LILLICRAP T，et al.Matching networks for one shot learning[C]//Neural Information Processing System，2016：3630-3638.
[8] SUNG F，YANG Y X，ZHANG L，et al.Learning to compare：relation network for few-shot learning[C]//Computer Vision and Pattern Recognition，2018.
[9] RUSSAKOVSKY O，DENG J，SU H，et al.Imagenet large scale visual recognition challenge[J].International Journal of Computer Vision，2015，115（3）：211-252.
[10] RAVI S，LAROCHELLE H.Optimization as a model for few-shot learning[C]//International Conference on Learning Representations，2017.
[11] FINN C，ABBEEL P，LEVINE S. Model-agnostic meta-learning for fast adaptation of deep networks[C]//International Conference on Machine Learning，2017：1126-1135.
[12] NICHOL A，ACHIAM J，SCHULMAN J.On first-order meta-learning algorithms[J].arXiv：1803.02999，2018.
[13] CHIH C C，CHIH J L.LIBSVM：a library for support vector machines[J].ACM Transactions on Intelligent Systems and Technology，2011，2（3）：1-39.
[14] HSU C，CHANG C，LIN C.A practical guide to support vector classification[J].BJU International，2008，101（1）：1396-1400.
[15] REN M，TRIANTAFILLOU E，RAVI S，et al.Meta-learning for semisupervised few-shot classification[J].arXiv：1803. 00676，2018.