Invariant Models for Causal Transfer Learning

被引:0
|
作者
Rojas-Carulla, Mateo [1 ,2 ]
Schoelkopf, Bernhard [1 ]
Turner, Richard [2 ]
Peters, Jonas [1 ,3 ]
机构
[1] Max Planck Inst Intelligent Syst, Tubingen, Germany
[2] Univ Cambridge, Dept Engn, Cambridge, England
[3] Univ Copenhagen, Dept Math Sci, Copenhagen, Denmark
基金
英国工程与自然科学研究理事会;
关键词
Transfer learning; Multi-task learning; Causality; Domain adaptation; Domain generalization; INFERENCE; NETWORKS;
D O I
暂无
中图分类号
TP [自动化技术、计算机技术];
学科分类号
0812 ;
摘要
Methods of transfer learning try to combine knowledge from several related tasks (or domains) to improve performance on a test task. Inspired by causal methodology, we relax the usual covariate shift assumption and assume that it holds true for a subset of predictor variables: the conditional distribution of the target variable given this subset of predictors is invariant over all tasks. We show how this assumption can be motivated from ideas in the field of causality. We focus on the problem of Domain Generalization, in which no examples from the test task are observed. We prove that in an adversarial setting using this subset for prediction is optimal in Domain Generalization; we further provide examples, in which the tasks are sufficiently diverse and the estimator therefore outperforms pooling the data, even on average. If examples from the test task are available, we also provide a method to transfer knowledge from the training tasks and exploit all available features for prediction. However, we provide no guarantees for this method. We introduce a practical method which allows for automatic inference of the above subset and provide corresponding code. We present results on synthetic data sets and a gene deletion data set.
引用
收藏
页数:34
相关论文
共 50 条
  • [31] Combining Reinforcement Learning and Causal Models for Robotics Applications
    Mendez-Molina, Arquimides
    PROCEEDINGS OF THE THIRTIETH INTERNATIONAL JOINT CONFERENCE ON ARTIFICIAL INTELLIGENCE, IJCAI 2021, 2021, : 4905 - 4906
  • [32] Learning linear cyclic causal models with latent variables
    Hyttinen, Antti
    Eberhardt, Frederick
    Hoyer, Patrik O.
    Journal of Machine Learning Research, 2012, 13 : 3387 - 3439
  • [33] TINET: Learning Invariant Networks via Knowledge Transfer
    Luo, Chen
    Chen, Zhengzhang
    Tang, Lu-An
    Shrivastava, Anshumali
    Li, Zhichun
    Chen, Haifeng
    Ye, Jieping
    KDD'18: PROCEEDINGS OF THE 24TH ACM SIGKDD INTERNATIONAL CONFERENCE ON KNOWLEDGE DISCOVERY & DATA MINING, 2018, : 1890 - 1899
  • [34] Active learning for optimal intervention design in causal models
    Jiaqi Zhang
    Louis Cammarata
    Chandler Squires
    Themistoklis P. Sapsis
    Caroline Uhler
    Nature Machine Intelligence, 2023, 5 : 1066 - 1075
  • [35] Learning Linear Cyclic Causal Models with Latent Variables
    Hyttinen, Antti
    Eberhardt, Frederick
    Hoyer, Patrik O.
    JOURNAL OF MACHINE LEARNING RESEARCH, 2012, 13 : 3387 - 3439
  • [36] Distributed learning of multi-agent causal models
    Meganck, S
    Maes, S
    Manderick, B
    Leray, P
    2005 IEEE/WIC/ACM INTERNATIONAL CONFERENCE ON INTELLIGENT AGENT TECHNOLOGY, PROCEEDINGS, 2005, : 285 - 288
  • [37] Learning the Structure of Causal Models with Relational and Temporal Dependence
    Marazopoulou, Katerina
    Maier, Marc
    Jensen, David
    UNCERTAINTY IN ARTIFICIAL INTELLIGENCE, 2015, : 572 - 581
  • [38] Challenges in Learning Causal Models of Alarms in Industrial Plants
    Wunderlich, Paul
    Niggemann, Oliver
    2018 IEEE 16TH INTERNATIONAL CONFERENCE ON INDUSTRIAL INFORMATICS (INDIN), 2018, : 623 - 628
  • [39] Causal graphical models with latent variables: Learning and inference
    Meganck, Stijn
    Leray, Philippe
    Manderick, Bernard
    SYMBOLIC AND QUANTITATIVE APPROACHES TO REASONING WITH UNCERTAINTY, PROCEEDINGS, 2007, 4724 : 5 - +
  • [40] GCISG: Guided Causal Invariant Learning for Improved Syn-to-Real Generalization
    Nam, Gilhyun
    Choi, Gyeongjae
    Lee, Kyungmin
    COMPUTER VISION - ECCV 2022, PT XXXIII, 2022, 13693 : 656 - 672