Skip to content

Commit d72b9c1

Browse files
fradavfradav
committed
Merge branch 'master' of github.com:computorg/computorg.github.io
2 parents 30f28af + 3cc9a19 commit d72b9c1

File tree

3 files changed

+174
-82
lines changed

3 files changed

+174
-82
lines changed

‎.gitignore‎

Lines changed: 2 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -15,4 +15,5 @@ vendor
1515
.env-secret
1616
.fake
1717
*_files/
18-
.vscode/
18+
.vscode/
19+
**/*.quarto_ipynb

‎site/about.qmd‎

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -46,6 +46,8 @@ The CC-BY license is therefore indicated on the published and readable versions
4646

4747
The source code of each article is based on various software and programs whose licenses (typically MIT, GPL, etc.) are specified where the code for this software is distributed. The authors commit to using content that complies with the CC-BY license of the articles and, more generally, to promoting the dissemination of open-source software, crediting the authors and contributors.
4848

49+
Computo does not ask for copyright transfer from authors: authors retain the copyright of their work.
50+
4951
This is in accordance with the [Budapest Open Access Initiative (BOAI)](https://en.wikipedia.org/wiki/Budapest_Open_Access_Initiative) definition of open access, as well as the [DOAJ](https://en.wikipedia.org/wiki/Directory_of_Open_Access_Journals)’s definition of open access, to whom we have submitted an application.
5052

5153
### Open reviews

‎site/published.yml‎

Lines changed: 170 additions & 81 deletions
Original file line numberDiff line numberDiff line change
@@ -1,48 +1,178 @@
1-
- abstract'@: >-
2-
This study investigates the use of Variational
3-
Auto-Encoders to build a simulator that approximates the law of
4-
genuine observations. Using both simulated and real data in
5-
scenarios involving counterfactuality, we discuss the general task
6-
of evaluating a simulator’s quality, with a focus on comparisons of
7-
statistical properties and predictive performance. While the
8-
simulator built from simulated data shows minor discrepancies, the
9-
results with real data reveal more substantial challenges. Beyond
10-
the technical analysis, we reflect on the broader implications of
11-
simulator design, and consider its role in modeling reality.
12-
authors@: Sandrine Boulet and Antoine Chambaz
13-
bibtex@: >+
14-
@article{boulet2025,
15-
author = {Boulet, Sandrine and Chambaz, Antoine},
1+
- abstract': >-
2+
In Bayesian statistics, the choice of the prior can have
3+
an important influence on the posterior and the parameter
4+
estimation, especially when few data samples are available. To limit
5+
the added subjectivity from a priori information, one can use the
6+
framework of objective priors, more particularly, we focus on
7+
reference priors in this work. However, computing such priors is a
8+
difficult task in general. Hence, we consider cases where the
9+
reference prior simplifies to the Jeffreys prior. We develop in this
10+
paper a flexible algorithm based on variational inference which
11+
computes approximations of priors from a set of parametric
12+
distributions using neural networks. We also show that our algorithm
13+
can retrieve modified Jeffreys priors when constraints are specified
14+
in the optimization problem to ensure the solution is proper. We
15+
propose a simple method to recover a relevant approximation of the
16+
parametric posterior distribution using Markov Chain Monte Carlo
17+
(MCMC) methods even if the density function of the parametric prior
18+
is not known in general. Numerical experiments on several
19+
statistical models of increasing complexity are presented. We show
20+
the usefulness of this approach by recovering the target
21+
distribution. The performance of the algorithm is evaluated on both
22+
prior and posterior distributions, jointly using variational
23+
inference and MCMC sampling.
24+
authors: Nils Baillie, Antoine Van Biesbroeck and Clément Gauchy
25+
bibtex: >+
26+
@article{baillie2025,
27+
author = {Baillie, Nils and Van Biesbroeck, Antoine and Gauchy,
28+
Clément},
1629
publisher = {French Statistical Society},
17-
title = {Draw {Me} a {Simulator}},
30+
title = {Variational Inference for Approximate Objective Priors Using
31+
Neural Networks},
1832
journal = {Computo},
19-
date = {2025-09-08},
20-
doi = {10.57750/w1hj-dw22},
33+
date = {2025-12-01},
34+
doi = {10.57750/76fh-t442},
2135
issn = {2824-7795},
2236
langid = {en},
23-
abstract = {This study investigates the use of Variational
24-
Auto-Encoders to build a simulator that approximates the law of
25-
genuine observations. Using both simulated and real data in
26-
scenarios involving counterfactuality, we discuss the general task
27-
of evaluating a simulator’s quality, with a focus on comparisons of
28-
statistical properties and predictive performance. While the
29-
simulator built from simulated data shows minor discrepancies, the
30-
results with real data reveal more substantial challenges. Beyond
31-
the technical analysis, we reflect on the broader implications of
32-
simulator design, and consider its role in modeling reality.}
37+
abstract = {In Bayesian statistics, the choice of the prior can have
38+
an important influence on the posterior and the parameter
39+
estimation, especially when few data samples are available. To limit
40+
the added subjectivity from a priori information, one can use the
41+
framework of objective priors, more particularly, we focus on
42+
reference priors in this work. However, computing such priors is a
43+
difficult task in general. Hence, we consider cases where the
44+
reference prior simplifies to the Jeffreys prior. We develop in this
45+
paper a flexible algorithm based on variational inference which
46+
computes approximations of priors from a set of parametric
47+
distributions using neural networks. We also show that our algorithm
48+
can retrieve modified Jeffreys priors when constraints are specified
49+
in the optimization problem to ensure the solution is proper. We
50+
propose a simple method to recover a relevant approximation of the
51+
parametric posterior distribution using Markov Chain Monte Carlo
52+
(MCMC) methods even if the density function of the parametric prior
53+
is not known in general. Numerical experiments on several
54+
statistical models of increasing complexity are presented. We show
55+
the usefulness of this approach by recovering the target
56+
distribution. The performance of the algorithm is evaluated on both
57+
prior and posterior distributions, jointly using variational
58+
inference and MCMC sampling.}
3359
}
3460
35-
date@: 2025-09-08
36-
description@: ''
37-
doi@: 10.57750/w1hj-dw22
38-
draft@: false
39-
journal@: Computo
40-
pdf@: ''
41-
repo@: published-202509-boulet-simulator
42-
title@: Draw Me a Simulator
43-
url@: ''
44-
year@: 2025
45-
abstract': >-
61+
date: 2025-12-01
62+
description: ''
63+
doi: 10.57750/76fh-t442
64+
draft: false
65+
journal: Computo
66+
pdf: ''
67+
repo: published-202512-baillie-varp
68+
title: Variational inference for approximate objective priors using neural networks
69+
url: ''
70+
year: 2025
71+
- abstract': >-
72+
The Maximum Mean Discrepancy (MMD) is a kernel-based
73+
metric widely used for nonparametric tests and estimation. Recently,
74+
it has also been studied as an objective function for parametric
75+
estimation, as it has been shown to yield robust estimators. We have
76+
implemented MMD minimization for parameter inference in a wide range
77+
of statistical models, including various regression models, within
78+
an `R` package called `regMMD`. This paper provides an introduction
79+
to the `regMMD` package. We describe the available kernels and
80+
optimization procedures, as well as the default settings. Detailed
81+
applications to simulated and real data are provided.
82+
authors: Pierre Alquier and Mathieu Gerber
83+
bibtex: >+
84+
@article{alquier2025,
85+
author = {Alquier, Pierre and Gerber, Mathieu},
86+
publisher = {French Statistical Society},
87+
title = {`regMMD`: An {`R`} Package for Parametric Estimation and
88+
Regression with Maximum Mean Discrepancy},
89+
journal = {Computo},
90+
date = {2025-11-18},
91+
doi = {10.57750/d6d1-gb09},
92+
issn = {2824-7795},
93+
langid = {en},
94+
abstract = {The Maximum Mean Discrepancy (MMD) is a kernel-based
95+
metric widely used for nonparametric tests and estimation. Recently,
96+
it has also been studied as an objective function for parametric
97+
estimation, as it has been shown to yield robust estimators. We have
98+
implemented MMD minimization for parameter inference in a wide range
99+
of statistical models, including various regression models, within
100+
an `R` package called `regMMD`. This paper provides an introduction
101+
to the `regMMD` package. We describe the available kernels and
102+
optimization procedures, as well as the default settings. Detailed
103+
applications to simulated and real data are provided.}
104+
}
105+
106+
date: 2025-11-18
107+
description: This document provides a complete introduction to the template based on the `regMMD` package for `R`, that implements minimum distance estimation in various parametric and regression models using the maximum mean discrepancy (MMD) metric.
108+
doi: 10.57750/d6d1-gb09
109+
draft: false
110+
journal: Computo
111+
pdf: ''
112+
repo: published-202511-alquier-regmmd
113+
title: '`regMMD`: an `R` package for parametric estimation and regression with maximum mean discrepancy'
114+
url: ''
115+
year: 2025
116+
- abstract': >-
117+
This paper presents a new algorithm (and an additional
118+
trick) that allows to compute fastly an entire curve of post hoc
119+
bounds for the False Discovery Proportion when the underlying bound
120+
\$V\^{}*\_\{\textbackslash mathfrak\{R\}\}\$ construction is based
121+
on a reference family \$\textbackslash mathfrak\{R\}\$ with a forest
122+
structure à la @MR4178188. By an entire curve, we mean the values
123+
\$V\^{}*\_\{\textbackslash mathfrak\{R\}\}(S\_1),\textbackslash
124+
dotsc,V\^{}*\_\{\textbackslash mathfrak\{R\}\}(S\_m)\$ computed on a
125+
path of increasing selection sets \$S\_1\textbackslash
126+
subsetneq\textbackslash dotsb\textbackslash subsetneq S\_m\$,
127+
\$\textbar S\_t\textbar=t\$. The new algorithm leverages the fact
128+
that going from \$S\_t\$ to \$S\_\{t+1\}\$ is done by adding only
129+
one hypothesis. Compared to a more naive approach, the new algorithm
130+
has a complexity in \$O(\textbar\textbackslash mathcal K\textbar
131+
m)\$ instead of \$O(\textbar\textbackslash mathcal K\textbar
132+
m\^{}2)\$, where \$\textbar\textbackslash mathcal K\textbar\$ is the
133+
cardinality of the family.
134+
authors: Guillermo Durand
135+
bibtex: >+
136+
@article{durand2025,
137+
author = {Durand, Guillermo},
138+
publisher = {French Statistical Society},
139+
title = {Fast Confidence Bounds for the False Discovery Proportion
140+
over a Path of Hypotheses},
141+
journal = {Computo},
142+
date = {2025-10-09},
143+
doi = {10.57750/efbs-ef14},
144+
issn = {2824-7795},
145+
langid = {en},
146+
abstract = {This paper presents a new algorithm (and an additional
147+
trick) that allows to compute fastly an entire curve of post hoc
148+
bounds for the False Discovery Proportion when the underlying bound
149+
\$V\^{}*\_\{\textbackslash mathfrak\{R\}\}\$ construction is based
150+
on a reference family \$\textbackslash mathfrak\{R\}\$ with a forest
151+
structure à la @MR4178188. By an entire curve, we mean the values
152+
\$V\^{}*\_\{\textbackslash mathfrak\{R\}\}(S\_1),\textbackslash
153+
dotsc,V\^{}*\_\{\textbackslash mathfrak\{R\}\}(S\_m)\$ computed on a
154+
path of increasing selection sets \$S\_1\textbackslash
155+
subsetneq\textbackslash dotsb\textbackslash subsetneq S\_m\$,
156+
\$\textbar S\_t\textbar=t\$. The new algorithm leverages the fact
157+
that going from \$S\_t\$ to \$S\_\{t+1\}\$ is done by adding only
158+
one hypothesis. Compared to a more naive approach, the new algorithm
159+
has a complexity in \$O(\textbar\textbackslash mathcal K\textbar
160+
m)\$ instead of \$O(\textbar\textbackslash mathcal K\textbar
161+
m\^{}2)\$, where \$\textbar\textbackslash mathcal K\textbar\$ is the
162+
cardinality of the family.}
163+
}
164+
165+
date: 2025-10-09
166+
description: ''
167+
doi: 10.57750/efbs-ef14
168+
draft: false
169+
journal: Computo
170+
pdf: ''
171+
repo: published-202510-durand-fast
172+
title: Fast confidence bounds for the false discovery proportion over a path of hypotheses
173+
url: ''
174+
year: 2025
175+
- abstract': >-
46176
This study investigates the use of Variational
47177
Auto-Encoders to build a simulator that approximates the law of
48178
genuine observations. Using both simulated and real data in
@@ -86,48 +216,7 @@
86216
title: Draw Me a Simulator
87217
url: ''
88218
year: 2025
89-
- abstract'@: >-
90-
Model-based clustering provides a principled way of
91-
developing clustering methods. We develop a new model-based
92-
clustering methods for count data. The method combines clustering
93-
and variable selection for improved clustering. The method is based
94-
on conditionally independent Poisson mixture models and Poisson
95-
generalized linear models. The method is demonstrated on simulated
96-
data and data from an ultra running race, where the method yields
97-
excellent clustering and variable selection performance.
98-
authors@: Julien Jacques and Thomas Brendan Murphy
99-
bibtex@: >+
100-
@article{jacques2025,
101-
author = {Jacques, Julien and Brendan Murphy, Thomas},
102-
publisher = {French Statistical Society},
103-
title = {Model-Based {Clustering} and {Variable} {Selection} for
104-
{Multivariate} {Count} {Data}},
105-
journal = {Computo},
106-
date = {2025-07-01},
107-
doi = {10.57750/6v7b-8483},
108-
issn = {2824-7795},
109-
langid = {en},
110-
abstract = {Model-based clustering provides a principled way of
111-
developing clustering methods. We develop a new model-based
112-
clustering methods for count data. The method combines clustering
113-
and variable selection for improved clustering. The method is based
114-
on conditionally independent Poisson mixture models and Poisson
115-
generalized linear models. The method is demonstrated on simulated
116-
data and data from an ultra running race, where the method yields
117-
excellent clustering and variable selection performance.}
118-
}
119-
120-
date@: 2025-07-01
121-
description@: ''
122-
doi@: 10.57750/6v7b-8483
123-
draft@: false
124-
journal@: Computo
125-
pdf@: ''
126-
repo@: published-202507-jacques-count-data
127-
title@: Model-Based Clustering and Variable Selection for Multivariate Count Data
128-
url@: ''
129-
year@: 2025
130-
abstract': >-
219+
- abstract': >-
131220
Model-based clustering provides a principled way of
132221
developing clustering methods. We develop a new model-based
133222
clustering methods for count data. The method combines clustering

0 commit comments

Comments
 (0)