《计算机应用》唯一官方网站 ›› 2022, Vol. 42 ›› Issue (3): 708-712.DOI: 10.11772/j.issn.1001-9081.2021040758

• 2021年中国计算机学会人工智能会议(CCFAI 2021) • 上一篇    

基于贝叶斯权函数的模型无关元学习算法

许仁杰, 刘宝弟, 张凯, 刘伟锋()   

  1. 中国石油大学(华东) 海洋与空间信息学院,青岛 266580
  • 收稿日期:2021-05-11 修回日期:2021-07-14 接受日期:2021-07-19 发布日期:2022-04-09 出版日期:2022-03-10
  • 通讯作者: 刘伟锋
  • 作者简介:许仁杰(1998—),男,山东青岛人,硕士研究生,主要研究方向:元学习、任务度量
    刘宝弟(1984—),男,山东济宁人,副教授,博士,CCF会员,主要研究方向:深度学习、图像处理
    张凯(1980—),男,四川南充人,教授,博士,主要研究方向:人工智能、机器学习、油藏数值模拟与油气田开发工程、机器学习、油田大数据分析;
  • 基金资助:
    国家自然科学基金资助项目(61671480);中国石油天然气集团公司重大科技项目(ZD2019?183?008);模式识别国家实验室开放项目(202000009)

Model agnostic meta learning algorithm based on Bayesian weight function

Renjie XU, Baodi LIU, Kai ZHANG, Weifeng LIU()   

  1. School of Oceanography and Spatial Information,China University of Petroleum (East China),Qingdao Shandong 266580,China
  • Received:2021-05-11 Revised:2021-07-14 Accepted:2021-07-19 Online:2022-04-09 Published:2022-03-10
  • Contact: Weifeng LIU
  • About author:XU Renjie, born in 1998, M. S. candidate. His research interests include meta learning, task measure.
    LIU Baodi, born in 1984, Ph. D., associate professor. His research interests include deep learning, image processing.
    ZHANG Kai, born in 1980, Ph. D., professor. His research interests include artificial intelligence, machine learning, reservoir numerical simulation and oil and gas field development engineering, machine learning, oilfield big data analysis.
  • Supported by:
    National Natural Science Foundation of China(61671480);Major Scientific and Technological Projects of CNPC(ZD2019-183-008);Open Project of National Laboratory of Pattern Recognition(202000009)

摘要:

模型无关的元学习(MAML)是一种多任务的元学习算法,能使用不同的模型,并快速地在不同任务之间进行适应,但MAML在训练速度与准确率上还亟待提高。从高斯随机过程的角度出发对MAML的原理进行分析,提出一种基于贝叶斯权函数的模型无关元学习(BW-MAML)算法,该权函数利用贝叶斯分析设计并用于损失的加权。训练过程中,BW-MAML将每次抽样的任务视为遵循高斯分布,根据贝叶斯分析计算不同任务在分布中的概率,并根据任务在分布中的概率判断该任务重要程度,再以此赋以不同的权重,从而提高每次梯度下降中信息的利用率。在Omniglot与Mini-ImageNet数据集上的小样本图像学习实验结果表明,通过增加贝叶斯权函数,BW-MAML的训练效果在6任务训练2 500步后,在Mini-ImageNet上的准确率比MAML的准确率最高提高了1.9个百分点,并且最终准确率比MAML平均提升了0.907个百分点;在Omniglot上的准确率也平均提升了0.199个百分点。

关键词: 贝叶斯分析, 高斯随机过程, 机器学习, 元学习, 小样本学习

Abstract:

As a multi-task meta learning algorithm, Model Agnostic Meta Learning (MAML) can use different models and adapt quickly to different tasks, but it still needs to be improved in terms of training speed and accuracy. The principle of MAML was analyzed from the perspective of Gaussian stochastic process, and a new Model Agnostic Meta Learning algorithm based on Bayesian Weight function (BW-MAML) was proposed, in which the weight was assigned by Bayesian analysis. In the training process of BW-MAML, each sampling task was regarded as following a Gaussian distribution, and the importance of the task was determined according to the probability of the task in the distribution, and then the weight was assigned according to the importance, thus improving the utilization of information in each gradient descent. The small sample image learning experimental results on Omniglot and Mini-ImageNet datasets show that by adding Bayesian weight function, for training effect of BW-MAML after 2500 step with 6 tasks, the accuracy of BW-MAML is at most 1.9 percentage points higher than that of MAML, and the final accuracy is 0.907 percentage points higher than that of MAML on Mini-ImageNet averagely; the accuracy of BW-MAML on Omniglot is also improved by up to 0.199 percentage points averagely.

Key words: Bayesian analysis, Gaussian stochastic process, machine learning, meta learning, few-shot learning

中图分类号: