Skip to content

Commit

Permalink
add mob to related work
Browse files Browse the repository at this point in the history
  • Loading branch information
cnellington committed Mar 26, 2024
1 parent c7944b9 commit 0fc9965
Show file tree
Hide file tree
Showing 2 changed files with 28 additions and 44 deletions.
59 changes: 21 additions & 38 deletions joss/paper.bib
Original file line number Diff line number Diff line change
Expand Up @@ -218,26 +218,6 @@ @article{kuijjer_estimating_2019
file = {Full Text:/Users/calebellington/Zotero/storage/MLXVELW9/Kuijjer et al. - 2019 - Estimating Sample-Specific Regulatory Networks.pdf:application/pdf},
}

@article{kolar_estimating_2010,
title = {Estimating time-varying networks},
volume = {4},
issn = {1932-6157},
url = {http://arxiv.org/abs/0812.5087},
doi = {10.1214/09-AOAS308},
abstract = {Stochastic networks are a plausible representation of the relational information among entities in dynamic systems such as living cells or social communities. While there is a rich literature in estimating a static or temporally invariant network from observation data, little has been done toward estimating time-varying networks from time series of entity attributes. In this paper we present two new machine learning methods for estimating time-varying networks, which both build on a temporally smoothed \$l\_1\$-regularized logistic regression formalism that can be cast as a standard convex-optimization problem and solved efficiently using generic solvers scalable to large networks. We report promising results on recovering simulated time-varying networks. For real data sets, we reverse engineer the latent sequence of temporally rewiring political networks between Senators from the US Senate voting records and the latent evolving regulatory networks underlying 588 genes across the life cycle of Drosophila melanogaster from the microarray time course.},
number = {1},
urldate = {2024-03-26},
journal = {The Annals of Applied Statistics},
author = {Kolar, Mladen and Song, Le and Ahmed, Amr and Xing, Eric P.},
month = mar,
year = {2010},
note = {arXiv:0812.5087 [q-bio, stat]},
keywords = {Quantitative Biology - Molecular Networks, Quantitative Biology - Quantitative Methods, Statistics - Applications, Statistics - Machine Learning, Statistics - Methodology},
annote = {Comment: Published in at http://dx.doi.org/10.1214/09-AOAS308 the Annals of Applied Statistics (http://www.imstat.org/aoas/) by the Institute of Mathematical Statistics (http://www.imstat.org)},
file = {arXiv Fulltext PDF:/Users/calebellington/Zotero/storage/9PS5EVN5/Kolar et al. - 2010 - Estimating time-varying networks.pdf:application/pdf},
}


@article{wang_bayesian_2022,
title = {Bayesian {Edge} {Regression} in {Undirected} {Graphical} {Models} to {Characterize} {Interpatient} {Heterogeneity} in {Cancer}},
volume = {117},
Expand All @@ -256,7 +236,6 @@ @article{wang_bayesian_2022
file = {Accepted Version:/Users/calebellington/Zotero/storage/A7P25CK7/Wang et al. - 2022 - Bayesian Edge Regression in Undirected Graphical M.pdf:application/pdf},
}


@article{parikh_treegl_2011,
title = {{TREEGL}: reverse engineering tree-evolving gene networks underlying developing biological lineages},
volume = {27},
Expand Down Expand Up @@ -294,23 +273,6 @@ @cs.cmu.edu.
pages = {i196--204},
}

@article{hothorn_partykit_2015,
title = {partykit: {A} {Modular} {Toolkit} for {Recursive} {Partytioning} in {R}},
volume = {16},
issn = {1533-7928},
shorttitle = {partykit},
url = {http://jmlr.org/papers/v16/hothorn15a.html},
abstract = {The R package partykit provides a flexible toolkit for learning, representing, summarizing, and visualizing a wide range of tree- structured regression and classification models. The functionality encompasses: (a) basic infrastructure for representing trees (inferred by any algorithm) so that unified print/plot/predict methods are available; (b) dedicated methods for trees with constant fits in the leaves (or terminal nodes) along with suitable coercion functions to create such trees (e.g., by rpart, RWeka, PMML); (c) a reimplementation of conditional inference trees (ctree, originally provided in the party package); (d) an extended reimplementation of model-based recursive partitioning (mob, also originally in party) along with dedicated methods for trees with parametric models in the leaves. Here, a brief overview of the package and its design is given while more detailed discussions of items (a)—(d) are available in vignettes accompanying the package.},
number = {118},
urldate = {2024-03-23},
journal = {Journal of Machine Learning Research},
author = {Hothorn, Torsten and Zeileis, Achim},
year = {2015},
keywords = {notion},
pages = {3905--3909},
file = {Full Text PDF:/Users/calebellington/Zotero/storage/S9N4P6K3/Hothorn and Zeileis - 2015 - partykit A Modular Toolkit for Recursive Partytio.pdf:application/pdf},
}


@inproceedings{kolar_sparsistent_2009,
title = {Sparsistent {Learning} of {Varying}-coefficient {Models} with {Structural} {Changes}},
Expand All @@ -324,3 +286,24 @@ @inproceedings{kolar_sparsistent_2009
year = {2009},
file = {Full Text PDF:/Users/calebellington/Zotero/storage/NUTZZVA9/Kolar et al. - 2009 - Sparsistent Learning of Varying-coefficient Models.pdf:application/pdf},
}


@article{zeileis_model-based_2008,
title = {Model-{Based} {Recursive} {Partitioning}},
volume = {17},
issn = {1061-8600},
url = {https://doi.org/10.1198/106186008X319331},
doi = {10.1198/106186008X319331},
abstract = {Recursive partitioning is embedded into the general and well-established class of parametric models that can be fitted using M-type estimators (including maximum likelihood). An algorithm for model-based recursive partitioning is suggested for which the basic steps are: (1) fit a parametric model to a dataset; (2) test for parameter instability over a set of partitioning variables; (3) if there is some overall parameter instability, split the model with respect to the variable associated with the highest instability; (4) repeat the procedure in each of the daughter nodes. The algorithm yields a partitioned (or segmented) parametric model that can be effectively visualized and that subject-matter scientists are used to analyzing and interpreting.},
number = {2},
urldate = {2024-03-26},
journal = {Journal of Computational and Graphical Statistics},
author = {Zeileis, Achim and Hothorn, Torsten and Hornik, Kurt},
month = jun,
year = {2008},
note = {Publisher: Taylor \& Francis
\_eprint: https://doi.org/10.1198/106186008X319331},
keywords = {Change points, Maximum likelihood, Parameter instability},
pages = {492--514},
file = {Submitted Version:/Users/calebellington/Zotero/storage/CNSH3FXF/Zeileis et al. - 2008 - Model-Based Recursive Partitioning.pdf:application/pdf},
}
13 changes: 7 additions & 6 deletions joss/paper.md
Original file line number Diff line number Diff line change
Expand Up @@ -66,11 +66,12 @@ With the flexibility of context-dependent parameters, each context-specific mode
# Statement of Need

“Personalized modeling” is a statistical method that has started to gain popularity in recent years for representing complex and heterogeneous systems exhibiting individual, sample-specific effects, such as those prevalent in complex diseases, financial markets, and social systems.
In its basic form: $x_i \sim P(X_i ;θ_i)$, where $i$ indexes a sample, $θ_i$ is the parameters defining the sample-specific distribution, and $x_i$ corresponds to the observation drawn from this sample-specific distribution, where understanding sample heterogeneity is equivalent to estimating data distributions with sample-specific parameters.
Disparate lineages of research have made contributions to this problem, typically by imposing a constraint on the variation of model parameters across all subjects [@hastie_varying-coefficient_1993; @fan_statistical_1999; @wang_bayesian_2022; @kolar_estimating_2010; @kolar_sparsistent_2009; @parikh_treegl_2011; @kuijjer_estimating_2019].
However, prior methods only model variation for a few continuous covariates [@hastie_varying-coefficient_1993; @fan_statistical_1999; @wang_bayesian_2022; @kolar_sparsistent_2009], known groups [@kolar_estimating_2010; @parikh_treegl_2011], or latent partitions [@hothorn_partykit_2015], or otherwise employ sample-left-out methods and are statistically underpowered [@kuijjer_estimating_2019].
In its basic form: $x_i \sim P(X_i ;θ_i)$, where $i$ indexes a sample, $θ_i$ is the parameters defining the sample-specific distribution, and $x_i$ corresponds to the observation drawn from this sample-specific distribution, where understanding sample heterogeneity is equivalent to estimating data distributions with sample-specific parameters.
Some methods, such as sample-left-out models [@kuijjer_estimating_2019], provide sample-specific estimators without additional information but lack desirable statistical properties such as the ability to generalize to new samples or test model performance on held-out data.
Due to the difficulty of estimating sample-specific parameters, most methods make use of side information, covariates or “context,” as an indicator of sample-to-sample variation [@hastie_varying-coefficient_1993; @fan_statistical_1999; @wang_bayesian_2022; @kolar_sparsistent_2009; @parikh_treegl_2011].
However, prior methods only permit model variation across a few continuous covariates [@hastie_varying-coefficient_1993; @fan_statistical_1999; @wang_bayesian_2022], or a small number of groups [@kolar_sparsistent_2009; @parikh_treegl_2011; @zeileis_model-based_2008], and do not scale to high-dimensional, continuous, and sample-specific variation.
Recently, contextual explanation networks (CENs) reframed the sample-specific parameter estimation problem as a more flexible and generalizable latent variable inference problem that can naturally leverage any additional contextual data pertinent to the study subjects [@al-shedivat_contextual_2020].
Contextualized Machine Learning is a generalization and abstraction of the CEN method, which provides a unified mathematical framework for inferring and estimating personalized models of heterogeneous and context-dependent systems [@lengerich_contextualized_2023].
Contextualized Machine Learning is a generalization of the CEN method, which provides a unified mathematical framework for inferring and estimating personalized models of heterogeneous and context-dependent systems [@lengerich_contextualized_2023].

Formally, given subject data $X = \{X_i\}_{i=1}^N$ and context data $C = \{C_i\}_{i=1}^N$ where $i$ indexes subjects, each observed only once, we can express the likelihood of all data in the form of
$$P(X,C) \propto \int_{\theta} d\theta P_M (X \mid \theta) P ( \theta \mid C)\,$$
Expand All @@ -81,8 +82,8 @@ Conveniently, $C$ can contain any multivariate or real features that are relevan

`Contextualized` implements this framework for key types of context encoders and sample-specific models, opening up new avenues for quantitative analysis of complex and heterogeeneous data, and simplifying the process of transforming this data into results with plug-and-play analysis tools. In particular, `Contextualized`:

1. **Unifies Past Modeling Frameworks:** `Contextualized` unifies modeling approaches for both homogeneous and heterogeneous data, including population models, varying-coefficient models [@hastie_varying-coefficient_1993; @fan_statistical_1999; @wang_bayesian_2022; @kolar_sparsistent_2009], and partition-based models such as regression trees [@hothorn_partykit_2015], sub-population models, (latent) mixture models, cluster-specific and group-specific models, tree-varying models[@parikh_treegl_2011], and time-varying models [@kolar_estimating_2010].
Additionally, `Contextualized` naturally falls back to these more traditional modeling frameworks when complex heterogeneity is not present.
1. **Unifies Modeling Frameworks:** `Contextualized` unifies modeling approaches for both homogeneous and heterogeneous data, including population models, varying-coefficient models [@hastie_varying-coefficient_1993; @fan_statistical_1999; @wang_bayesian_2022], and partition-based models [@kolar_sparsistent_2009; @parikh_treegl_2011; @zeileis_model-based_2008] via context encoding, learning parameter variation over both continuous contexts and discrete groups.
Additionally, `Contextualized` naturally falls back to these classic modeling frameworks when complex heterogeneity is not present.
Not only is this convenient, but it limits the number of modeling decisions and validation tests required by users, reducing the risk of misspecification and false discoveries [@lengerich_contextualized_2023].
2. **Models High-resolution Heterogeneity:** Contextualized models adapt to the context of each sample by using a context encoder, naturally accounting for high-dimensional, continuous, and fine-grained variation between samples [@ellington_contextualized_2023].
3. **Quantifies Heterogeneity in Data:** Context-specific models quantify the randomness and structure of the systems underlying each data point, and variation in context-specific model parameters quantifies the heterogeneity between data points [@deuschel_contextualized_2023; @al-shedivat_personalized_2018].
Expand Down

0 comments on commit 0fc9965

Please sign in to comment.