學術產出-Theses
Article View/Open
Publication Export
-
題名 考量效果風險的處方配置：高斯過程迴歸
Treatment Allocation Subject to Effect Risks: A Gaussian Process Regression Approach作者 黃凱文
Huang, Kai-Wen貢獻者 莊皓鈞<br>周彥君
Chuang, Hao-Chun<br>Chou, Yen-Chun
黃凱文
Huang, Kai-Wen關鍵詞 高斯過程迴歸
處方分配
Gaussian Process Regression
Treatment Allocation日期 2021 上傳時間 4-Aug-2021 14:47:02 (UTC+8) 摘要 處方資源的分配是各領域重要的研究問題之一，而現代機器學習技術在此方面也有不少的進展，許多研究者利用不同的模型去估計處方效果，並以此做為處方配置的依據。然而，過去的處方效果的研究大多以點估計的方式進行，此舉並未考慮不同個體可能會產生異質變異數(heteroskedasticity)的處方效果，導致決策者在配置處方的過程中，無法根據不同的風險情況進行決策。因此，我們提出考慮個體風險的潛在結果框架，以潛在結果的區間估計，推估個體處方效果及其風險；其中，我們提出可以處理異質變異數的高斯過程(Heteroskedastic Gaussian Process, HGP)作為預測模型，其在保有高斯過程預測值具有概率性的優點下，也能處理個體間具有異質變異數的資料情況。而在模擬資料的實驗中，HGP也確實能在高度異質變異數的資料情況下準確地估計處方效果及效果風險，而其能根據不同的風險情況下彈性地改變分配決策，達到比一般高斯過程或隨機森林模型更好的分配準確度表現。本研究在處方分配領域提供新的分析途徑，並提出了適合估計效果風險預測模型，期望對風險下的處方分配問題有所貢獻。
The allocation of treatment resources is one of the important research issues in various fields, and modern machine learning technology has also made a lot of progress in this regard. Many researchers used different models to estimate treatment effects and used these as the basis for treatment allocations. However, most of the researches in the past estimated treatment effects by point-estimation. They did not consider that different individuals may have heterogeneous variance (heteroskedasticity) treatment effects, resulting in that decision makers can not make appropriate decisions under different risk situations.Therefore, we propose a potential outcome framework that considers risks, and use the interval estimation of potential outcomes to estimate the effects of individual treatment effects and their risks , and then we propose Heteroskedastic Gaussian Process (HGP) as the prediction model. HGP can handle data with heterogeneous variance between individuals while maintaining Gaussian process advantage that its prediction is probabilistic. In experiments with simulated data, HGP can indeed perform well in the case of highly heterogeneous variance data compared to normal GP and Random Forest model. This research provides a new approach of treatments allocation problem, and also proposes a model suitable for estimating heteroskedastic treatment effects. By these, we expect to make a contribution to the field of treatment allocations under risk.參考文獻 Athey, S., & Imbens, G. (2016). Recursive partitioning for heterogeneous causal effects. Proceedings of the National Academy of Sciences, 113(27), 7353-7360.Bertsimas, D., Dunn, J., & Mundru, N. (2019). Optimal prescriptive trees. INFORMS Journal on Optimization, 1(2), 164-183.Cochran, W. G., & Chambers, S. P. (1965). The planning of observational studies of human populations. Journal of the Royal Statistical Society. Series A (General), 128(2), 234-266.Deming, W. E. (1975). The logic of evaluation. Handbook of evaluation research, 1, 53-68.Duvenaud, D. (2014). Automatic model construction with Gaussian processes (Doctoral dissertation, University of Cambridge).Hensman, J., Fusi, N., & Lawrence, N. D. (2013). Gaussian processes for big data. arXiv preprint arXiv:1309.6835.Imbens, G. W., & Rubin, D. B. (2015). Causal inference in statistics, social, and biomedical sciences. Cambridge University Press.Kochenderfer, M. J., & Wheeler, T. A. (2019). Algorithms for optimization. Mit Press.Kube, A., Das, S., & Fowler, P. J. (2019). Allocating interventions based on predicted outcomes: A case study on homelessness services. Proceedings of the AAAI Conference on Artificial Intelligence, 33(01), 622-629.Lu, M., Sadiq, S., Feaster, D. J., & Ishwaran, H. (2018). Estimating individual treatment effect in observational data using random forest methods. Journal of Computational and Graphical Statistics, 27(1), 209-219.Rasmussen, C. E., & Williams, C. (2006). Gaussian processes for machine learning, vol. 1. In: MIT press Cambridge MA.Rosenbaum, P. R. (2005). Observational study. Encyclopedia of statistics in behavioral science, 3, 1451–1462.Rosenbaum, P. R., & Rubin, D. B. (1983). The central role of the propensity score in observational studies for causal effects. Biometrika, 70(1), 41-55.Rubin, D. B. (1974). Estimating causal effects of treatments in randomized and nonrandomized studies. Journal of educational Psychology, 66(5), 688.Schulz, E., Speekenbrink, M., & Krause, A. (2018). A tutorial on Gaussian process regression: Modelling, exploring, and exploiting functions. Journal of Mathematical Psychology, 85, 1-16.Titsias, M. (2009). Variational learning of inducing variables in sparse Gaussian processes. Artificial intelligence and statistics, 5, 567-576.Yao, L., Chu, Z., Li, S., Li, Y., Gao, J., & Zhang, A. (2020). A survey on causal inference. arXiv preprint arXiv:2002.02770.Yu, H., Xie, T., Paszczynski, S., & Wilamowski, B. M. (2011). Advantages of radial basis function networks for dynamic system design. IEEE Transactions on Industrial Electronics, 58(12), 5438-5450.何志昆, 刘光斌, 赵曦晶, & 王明昊. (2013). 高斯过程回归方法综述. 控制与决策, 28(8), 1121-1129. 描述 碩士
國立政治大學
資訊管理學系
108356007資料來源 http://thesis.lib.nccu.edu.tw/record/#G0108356007 資料類型 thesis dc.contributor.advisor 莊皓鈞<br>周彥君 zh_TW dc.contributor.advisor Chuang, Hao-Chun<br>Chou, Yen-Chun en_US dc.contributor.author (Authors) 黃凱文 zh_TW dc.contributor.author (Authors) Huang, Kai-Wen en_US dc.creator (作者) 黃凱文 zh_TW dc.creator (作者) Huang, Kai-Wen en_US dc.date (日期) 2021 en_US dc.date.accessioned 4-Aug-2021 14:47:02 (UTC+8) - dc.date.available 4-Aug-2021 14:47:02 (UTC+8) - dc.date.issued (上傳時間) 4-Aug-2021 14:47:02 (UTC+8) - dc.identifier (Other Identifiers) G0108356007 en_US dc.identifier.uri (URI) http://nccur.lib.nccu.edu.tw/handle/140.119/136341 - dc.description (描述) 碩士 zh_TW dc.description (描述) 國立政治大學 zh_TW dc.description (描述) 資訊管理學系 zh_TW dc.description (描述) 108356007 zh_TW dc.description.abstract (摘要) 處方資源的分配是各領域重要的研究問題之一，而現代機器學習技術在此方面也有不少的進展，許多研究者利用不同的模型去估計處方效果，並以此做為處方配置的依據。然而，過去的處方效果的研究大多以點估計的方式進行，此舉並未考慮不同個體可能會產生異質變異數(heteroskedasticity)的處方效果，導致決策者在配置處方的過程中，無法根據不同的風險情況進行決策。因此，我們提出考慮個體風險的潛在結果框架，以潛在結果的區間估計，推估個體處方效果及其風險；其中，我們提出可以處理異質變異數的高斯過程(Heteroskedastic Gaussian Process, HGP)作為預測模型，其在保有高斯過程預測值具有概率性的優點下，也能處理個體間具有異質變異數的資料情況。而在模擬資料的實驗中，HGP也確實能在高度異質變異數的資料情況下準確地估計處方效果及效果風險，而其能根據不同的風險情況下彈性地改變分配決策，達到比一般高斯過程或隨機森林模型更好的分配準確度表現。本研究在處方分配領域提供新的分析途徑，並提出了適合估計效果風險預測模型，期望對風險下的處方分配問題有所貢獻。 zh_TW dc.description.abstract (摘要) The allocation of treatment resources is one of the important research issues in various fields, and modern machine learning technology has also made a lot of progress in this regard. Many researchers used different models to estimate treatment effects and used these as the basis for treatment allocations. However, most of the researches in the past estimated treatment effects by point-estimation. They did not consider that different individuals may have heterogeneous variance (heteroskedasticity) treatment effects, resulting in that decision makers can not make appropriate decisions under different risk situations.Therefore, we propose a potential outcome framework that considers risks, and use the interval estimation of potential outcomes to estimate the effects of individual treatment effects and their risks , and then we propose Heteroskedastic Gaussian Process (HGP) as the prediction model. HGP can handle data with heterogeneous variance between individuals while maintaining Gaussian process advantage that its prediction is probabilistic. In experiments with simulated data, HGP can indeed perform well in the case of highly heterogeneous variance data compared to normal GP and Random Forest model. This research provides a new approach of treatments allocation problem, and also proposes a model suitable for estimating heteroskedastic treatment effects. By these, we expect to make a contribution to the field of treatment allocations under risk. en_US dc.description.tableofcontents 第一章 緒論 1第二章 文獻探討 4 第一節 因果推論 4 第二節 潛在結果框架與處方分配 5 第三節 高斯過程迴歸 7第三章 研究方法 11 第一節 處方分配問題與選擇準則 11 第二節 風險下的潛在結果框架 13 第三節 處方效果估計與模型評估 15 第四節 異質變異數高斯過程迴歸 17第四章 實驗與結果 21 第一節 實驗設計 21 第二節 結果分析─處方分配問題 27 第三節 參數分析 33第五章 結論 37第六章 參考文獻 38 zh_TW dc.format.extent 1417904 bytes - dc.format.mimetype application/pdf - dc.source.uri (資料來源) http://thesis.lib.nccu.edu.tw/record/#G0108356007 en_US dc.subject (關鍵詞) 高斯過程迴歸 zh_TW dc.subject (關鍵詞) 處方分配 zh_TW dc.subject (關鍵詞) Gaussian Process Regression en_US dc.subject (關鍵詞) Treatment Allocation en_US dc.title (題名) 考量效果風險的處方配置：高斯過程迴歸 zh_TW dc.title (題名) Treatment Allocation Subject to Effect Risks: A Gaussian Process Regression Approach en_US dc.type (資料類型) thesis en_US dc.relation.reference (參考文獻) Athey, S., & Imbens, G. (2016). Recursive partitioning for heterogeneous causal effects. Proceedings of the National Academy of Sciences, 113(27), 7353-7360.Bertsimas, D., Dunn, J., & Mundru, N. (2019). Optimal prescriptive trees. INFORMS Journal on Optimization, 1(2), 164-183.Cochran, W. G., & Chambers, S. P. (1965). The planning of observational studies of human populations. Journal of the Royal Statistical Society. Series A (General), 128(2), 234-266.Deming, W. E. (1975). The logic of evaluation. Handbook of evaluation research, 1, 53-68.Duvenaud, D. (2014). Automatic model construction with Gaussian processes (Doctoral dissertation, University of Cambridge).Hensman, J., Fusi, N., & Lawrence, N. D. (2013). Gaussian processes for big data. arXiv preprint arXiv:1309.6835.Imbens, G. W., & Rubin, D. B. (2015). Causal inference in statistics, social, and biomedical sciences. Cambridge University Press.Kochenderfer, M. J., & Wheeler, T. A. (2019). Algorithms for optimization. Mit Press.Kube, A., Das, S., & Fowler, P. J. (2019). Allocating interventions based on predicted outcomes: A case study on homelessness services. Proceedings of the AAAI Conference on Artificial Intelligence, 33(01), 622-629.Lu, M., Sadiq, S., Feaster, D. J., & Ishwaran, H. (2018). Estimating individual treatment effect in observational data using random forest methods. Journal of Computational and Graphical Statistics, 27(1), 209-219.Rasmussen, C. E., & Williams, C. (2006). Gaussian processes for machine learning, vol. 1. In: MIT press Cambridge MA.Rosenbaum, P. R. (2005). Observational study. Encyclopedia of statistics in behavioral science, 3, 1451–1462.Rosenbaum, P. R., & Rubin, D. B. (1983). The central role of the propensity score in observational studies for causal effects. Biometrika, 70(1), 41-55.Rubin, D. B. (1974). Estimating causal effects of treatments in randomized and nonrandomized studies. Journal of educational Psychology, 66(5), 688.Schulz, E., Speekenbrink, M., & Krause, A. (2018). A tutorial on Gaussian process regression: Modelling, exploring, and exploiting functions. Journal of Mathematical Psychology, 85, 1-16.Titsias, M. (2009). Variational learning of inducing variables in sparse Gaussian processes. Artificial intelligence and statistics, 5, 567-576.Yao, L., Chu, Z., Li, S., Li, Y., Gao, J., & Zhang, A. (2020). A survey on causal inference. arXiv preprint arXiv:2002.02770.Yu, H., Xie, T., Paszczynski, S., & Wilamowski, B. M. (2011). Advantages of radial basis function networks for dynamic system design. IEEE Transactions on Industrial Electronics, 58(12), 5438-5450.何志昆, 刘光斌, 赵曦晶, & 王明昊. (2013). 高斯过程回归方法综述. 控制与决策, 28(8), 1121-1129. zh_TW dc.identifier.doi (DOI) 10.6814/NCCU202100981 en_US
