Update notes

Author: cxy0714

notes/Master.bib

@@ -1,59 +1,85 @@
+@article{barber2015controlling,
+  title  = {Controlling the false discovery rate via knockoffs},
+  author = {Barber, Rina Foygel and Cand{\`e}s, Emmanuel J},
+  year   = {2015}
+}
+@article{ren2024derandomised,
+  title     = {Derandomised knockoffs: leveraging e-values for false discovery rate control},
+  author    = {Ren, Zhimei and Barber, Rina Foygel},
+  journal   = {Journal of the Royal Statistical Society Series B: Statistical Methodology},
+  volume    = {86},
+  number    = {1},
+  pages     = {122--154},
+  year      = {2024},
+  publisher = {Oxford University Press US}
+}
+@article{zrnic2023post,
+  title={Post-selection inference via algorithmic stability},
+  author={Zrnic, Tijana and Jordan, Michael I},
+  journal={The Annals of Statistics},
+  volume={51},
+  number={4},
+  pages={1666--1691},
+  year={2023},
+  publisher={Institute of Mathematical Statistics}
+}
+
 @article{lin2025role,
-  title={The role of nudges in food choices: An umbrella review},
-  author={Lin, Hui and de Barcellos, Marcia Dutra and De Steur, Hans},
-  journal={Food Quality and Preference},
-  pages={105679},
-  year={2025},
-  publisher={Elsevier}
+  title     = {The role of nudges in food choices: An umbrella review},
+  author    = {Lin, Hui and de Barcellos, Marcia Dutra and De Steur, Hans},
+  journal   = {Food Quality and Preference},
+  pages     = {105679},
+  year      = {2025},
+  publisher = {Elsevier}
 }
 
 @article{shea2017amstar,
-  title={AMSTAR 2: a critical appraisal tool for systematic reviews that include randomised or non-randomised studies of healthcare interventions, or both},
-  author={Shea, Beverley J and Reeves, Barnaby C and Wells, George and Thuku, Micere and Hamel, Candyce and Moran, Julian and Moher, David and Tugwell, Peter and Welch, Vivian and Kristjansson, Elizabeth and others},
-  journal={bmj},
-  volume={358},
-  year={2017},
-  publisher={British Medical Journal Publishing Group}
+  title     = {AMSTAR 2: a critical appraisal tool for systematic reviews that include randomised or non-randomised studies of healthcare interventions, or both},
+  author    = {Shea, Beverley J and Reeves, Barnaby C and Wells, George and Thuku, Micere and Hamel, Candyce and Moran, Julian and Moher, David and Tugwell, Peter and Welch, Vivian and Kristjansson, Elizabeth and others},
+  journal   = {bmj},
+  volume    = {358},
+  year      = {2017},
+  publisher = {British Medical Journal Publishing Group}
 }
 @inproceedings{jamshidi2020peregrine,
-  title={Peregrine: a pattern-aware graph mining system},
-  author={Jamshidi, Kasra and Mahadasa, Rakesh and Vora, Keval},
-  booktitle={Proceedings of the Fifteenth European Conference on Computer Systems},
-  pages={1--16},
-  year={2020}
+  title     = {Peregrine: a pattern-aware graph mining system},
+  author    = {Jamshidi, Kasra and Mahadasa, Rakesh and Vora, Keval},
+  booktitle = {Proceedings of the Fifteenth European Conference on Computer Systems},
+  pages     = {1--16},
+  year      = {2020}
 }
 @inproceedings{mawhirter2019automine,
-  title={Automine: harmonizing high-level abstraction and high performance for graph mining},
-  author={Mawhirter, Daniel and Wu, Bo},
-  booktitle={Proceedings of the 27th ACM Symposium on Operating Systems Principles},
-  pages={509--523},
-  year={2019}
+  title     = {Automine: harmonizing high-level abstraction and high performance for graph mining},
+  author    = {Mawhirter, Daniel and Wu, Bo},
+  booktitle = {Proceedings of the 27th ACM Symposium on Operating Systems Principles},
+  pages     = {509--523},
+  year      = {2019}
 }
 
 @inproceedings{wang2016gunrock,
-  title={Gunrock: A high-performance graph processing library on the GPU},
-  author={Wang, Yangzihao and Davidson, Andrew and Pan, Yuechao and Wu, Yuduo and Riffel, Andy and Owens, John D},
-  booktitle={Proceedings of the 21st ACM SIGPLAN symposium on principles and practice of parallel programming},
-  pages={1--12},
-  year={2016}
+  title     = {Gunrock: A high-performance graph processing library on the GPU},
+  author    = {Wang, Yangzihao and Davidson, Andrew and Pan, Yuechao and Wu, Yuduo and Riffel, Andy and Owens, John D},
+  booktitle = {Proceedings of the 21st ACM SIGPLAN symposium on principles and practice of parallel programming},
+  pages     = {1--12},
+  year      = {2016}
 }
 
 @incollection{fender2022rapids,
-  title={Rapids cugraph},
-  author={Fender, Alex and Rees, Brad and Eaton, Joe},
-  booktitle={Massive Graph Analytics},
-  pages={483--493},
-  year={2022},
-  publisher={Chapman and Hall/CRC}
+  title     = {Rapids cugraph},
+  author    = {Fender, Alex and Rees, Brad and Eaton, Joe},
+  booktitle = {Massive Graph Analytics},
+  pages     = {483--493},
+  year      = {2022},
+  publisher = {Chapman and Hall/CRC}
 }
 
 @article{levis2025covariate,
-  title={Covariate-assisted bounds on causal effects with instrumental variables},
-  author={Levis, Alexander W and Bonvini, Matteo and Zeng, Zhenghao and Keele, Luke and Kennedy, Edward H},
-  journal={Journal of the Royal Statistical Society Series B: Statistical Methodology},
-  pages={qkaf028},
-  year={2025},
-  publisher={Oxford University Press UK}
+  title     = {Covariate-assisted bounds on causal effects with instrumental variables},
+  author    = {Levis, Alexander W and Bonvini, Matteo and Zeng, Zhenghao and Keele, Luke and Kennedy, Edward H},
+  journal   = {Journal of the Royal Statistical Society Series B: Statistical Methodology},
+  pages     = {qkaf028},
+  year      = {2025},
+  publisher = {Oxford University Press UK}
 }
 
 @article{wu2025promises,

notes/main.typ

@@ -224,7 +224,7 @@ Here $serif(Pr)(Y(1) = b | G = g)$ and $serif(Pr)( Y(0) = a | G = g)$ can be ide
 
 === Numerical calcaulation of influence function
 @mukhinkernel
-
+@jordan2022empirical
 === Von mise representation
 
 === Tangent space
@@ -257,9 +257,33 @@ Formulation in @van1991differentiable
 
 See #link("https://statmodeling.stat.columbia.edu/2025/11/12/belief-in-the-law-of-small-numbers-as-a-way-to-understand-the-replication-crisis-and-silly-researchers-who-continue-to-cite-discredited-behavioral-research/")[`Belief in the law of small numbers as a way to understand the replication crisis and silly researchers who continue to cite discredited behavioral research`] by Andrew Gelman, but I just see the paper involved a title "umbrella review" @lin2025role, so I search more about umbrella review.
 
-Umbrella review consider the evidence form multiple meta-analysis and system review studies on the same topic. 
+Umbrella review consider the evidence form multiple meta-analysis and system review studies on the same topic.
 
 using @shea2017amstar(10K+ citation!) to assess the quality of included meta-analysis/system review studies, but they did not provide a way to aggregate the quality scores and just give a subjective criteria, the overall confidence is rating as "high", "moderate", "low", "critically low". Produce an conclusion such that "In the 60 meta-analyses/systematic reviews included on the treatment for type 2 diabetes, treatments A have 5 strong evidence to support its effectiveness, while treatment B has only 1 moderate evidence to support its effectiveness..."
 
+= multiple testing
+
+== Varibale selection
+
+A talk given by Zhong Wei: #link("https://www.cmstatistics.org/RegistrationsV2/EcoSta2025/viewSubmission.php?in=1158&token=472n51nsro843q0p0s3sn1663po5rsrr")[` A unified stability approach to false discovery rate control`]
+
+Knockoff is a randomized method for variable selection controlling false discovery rate(FDR) @barber2015controlling. The key idea is to construct knockoff variables $tilde(X) in R^p$ that mimic the correlation structure of the original variables $X in R^p$, such that for any subset $S subset {1, ..., p}$, swapping the variables in $S$ with their knockoffs does not change the joint distribution of $(X, tilde(X))$, while also ensuring that the knockoff variables are conditionally independent of the response variable $Y$ given the original variables $X$.
+
+randomized method have a problem in reproduciblity, e-value developed by @vovk2021values is a way to solve this problem, the e-value is a function of data that has an expected value at most 1 under the null hypothesis, thus can be used to control type I error in multiple testing, it can combine evidence from multiple independent tests easily, by running the knockoff procedure multiple times and averaging the resulting e-values for each variable, we can obtain a more stable measure of evidence against the null hypothesis for each variable, that's the work in @ren2024derandomised.
+
+Zhong Wei proposed a general framework of stability for FDR control, which includes derandomized knockoff as a special case, and provide theoretical guarantee for FDR control under this framework.
+
+Their another work is consider the inference after variable selection, using the same sample for variable selection and inference will lead to selection bias @zrnic2023post.
+
+#question(
+  "Criteria for variable selection controlling FDR",
+)[A quesition is deose there have some minimax optimal method or other criteria for variable selection controlling FDR? e-value BH method or p-value BH method? or here derandomized knockoff BH method? Which is best?]
+
+#link("https://rinafb.github.io/")[Rina Foygel Barber]是2025年新任美国科学院院士，今年统计方向的两位美国科学院（National Academy of Sciences）院士，另一位是刘军，Rina Foygel Barber是2020年COPSS总统奖（12年phd毕业 at U of Chicago，postdoc under Emmanuel Candès）获得者，颁奖理由是：
+
+#quote(
+  "For fundamental contributions to statistical sparsity and selective inference in high-dimensional problems, for the creative and novel knockoff filter to cope with correlated coefficients, for contributions to compressed sensing, the jackknife, and conformal predictive inference; for the encouragement and training of graduate and undergraduate students.",
+)
 
+e-valuede 的#link("https://sas.uwaterloo.ca/~wang/")[王若度]（U of Waterloo，  Chair profess） 曾经是星际争霸职业选手。
 #bibliography("Master.bib")