You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
bio: "Michele is a PhD student researching randomized and trainable models for time series and graphs. The focus of is research is on uncertainty quantification and interpretability."
- title: "MaxCutPool: differentiable feature-aware Maxcut for pooling in graph neural networks"
2
+
authors: "Carlo Abate, Filippo Maria Bianchi"
3
+
figure: "figs/publications/maxcutpool.png"
4
+
abstract: "We propose a novel approach to compute the MAXCUT in attributed graphs, i.e., graphs with features associated with nodes and edges. Our approach is robust to the underlying graph topology and is fully differentiable, making it possible to find solutions that jointly optimize the MAXCUT along with other objectives. Based on the obtained MAXCUT partition, we implement a hierarchical graph pooling layer for Graph Neural Networks, which is sparse, differentiable, and particularly suitable for downstream tasks on heterophilic graphs."
5
+
github: "https://github.com/NGMLGroup/MaxCutPool"
6
+
arxiv: "https://arxiv.org/abs/2409.05100"
7
+
bibtex: |
8
+
@article{abate2024maxcutpool,
9
+
title={MaxCutPool: differentiable feature-aware Maxcut for pooling in graph neural networks},
10
+
author={Abate, Carlo and Bianchi, Filippo Maria},
11
+
journal={arXiv preprint arXiv:2409.05100},
12
+
year={2024}
13
+
}
14
+
1
15
- title: "Graph-based Forecasting with Missing Data through Spatiotemporal Downsampling"
2
16
authors: "Ivan Marisca, Cesare Alippi, Filippo Maria Bianchi"
3
17
venue: "ICML 2024"
@@ -15,18 +29,4 @@
15
29
volume = {235},
16
30
series = {Proceedings of Machine Learning Research},
17
31
publisher = {PMLR}
18
-
}
19
-
20
-
- title: "MaxCutPool: differentiable feature-aware Maxcut for pooling in graph neural networks"
21
-
authors: "Carlo Abate, Filippo Maria Bianchi"
22
-
figure: "figs/publications/maxcutpool.png"
23
-
abstract: "We propose a novel approach to compute the MAXCUT in attributed graphs, i.e., graphs with features associated with nodes and edges. Our approach is robust to the underlying graph topology and is fully differentiable, making it possible to find solutions that jointly optimize the MAXCUT along with other objectives. Based on the obtained MAXCUT partition, we implement a hierarchical graph pooling layer for Graph Neural Networks, which is sparse, differentiable, and particularly suitable for downstream tasks on heterophilic graphs."
24
-
github: "https://github.com/NGMLGroup/MaxCutPool"
25
-
arxiv: "https://arxiv.org/abs/2409.05100"
26
-
bibtex: |
27
-
@article{abate2024maxcutpool,
28
-
title={MaxCutPool: differentiable feature-aware Maxcut for pooling in graph neural networks},
<?xml version="1.0" encoding="utf-8"?><feedxmlns="http://www.w3.org/2005/Atom" ><generatoruri="https://jekyllrb.com/"version="4.3.4">Jekyll</generator><linkhref="http://localhost:4000/feed.xml"rel="self"type="application/atom+xml" /><linkhref="http://localhost:4000/"rel="alternate"type="text/html" /><updated>2024-10-14T12:39:42+02:00</updated><id>http://localhost:4000/feed.xml</id><titletype="html">Northernmost GraphML Group</title><subtitle>The Northermost GraphML group in the world, based in Tromsø, Norway.
1
+
<?xml version="1.0" encoding="utf-8"?><feedxmlns="http://www.w3.org/2005/Atom" ><generatoruri="https://jekyllrb.com/"version="4.3.4">Jekyll</generator><linkhref="http://localhost:4000/feed.xml"rel="self"type="application/atom+xml" /><linkhref="http://localhost:4000/"rel="alternate"type="text/html" /><updated>2024-10-14T13:44:30+02:00</updated><id>http://localhost:4000/feed.xml</id><titletype="html">Northernmost GraphML Group</title><subtitle>The Northermost GraphML group in the world, based in Tromsø, Norway.
Copy file name to clipboardExpand all lines: _site/publications.html
+31-32Lines changed: 31 additions & 32 deletions
Original file line number
Diff line number
Diff line change
@@ -55,39 +55,35 @@ <h1>Publications</h1>
55
55
<divclass="publication">
56
56
57
57
<!-- Title -->
58
-
<h3class="pub-title"><strong>Graph-based Forecasting with Missing Data through Spatiotemporal Downsampling</strong></h3>
58
+
<h3class="pub-title"><strong>MaxCutPool: differentiable feature-aware Maxcut for pooling in graph neural networks</strong></h3>
59
59
60
60
<!-- Authors -->
61
61
<pclass="authors">
62
-
<emstyle="color: gray;">Ivan Marisca, Cesare Alippi, Filippo Maria Bianchi</em>
62
+
<emstyle="color: gray;">Carlo Abate, Filippo Maria Bianchi</em>
63
63
</p>
64
64
65
65
<!-- Venue (optional) -->
66
66
67
-
<pclass="venue">
68
-
<em>ICML 2024</em>
69
-
</p>
70
-
71
67
72
68
<!-- Figure -->
73
69
74
70
<divclass="pub-figure">
75
-
<imgsrc="figs/publications/hdtts.png" alt="Figure for Graph-based Forecasting with Missing Data through Spatiotemporal Downsampling">
71
+
<imgsrc="figs/publications/maxcutpool.png" alt="Figure for MaxCutPool: differentiable feature-aware Maxcut for pooling in graph neural networks">
76
72
</div>
77
73
78
74
79
75
<!-- Abstract -->
80
-
<p>Spatiotemporal graph neural networks achieve striking results by representing the relationships across time series as a graph. Nonetheless, most existing methods rely on the often unrealistic assumption that inputs are always available and fail to capture hidden spatiotemporal dynamics when part of the data is missing. In this work, we tackle this problem through hierarchical spatiotemporal downsampling. The input time series are progressively coarsened over time and space, obtaining a pool of representations that capture heterogeneous temporal and spatial dynamics. Conditioned on observations and missing data patterns, such representations are combined by an interpretable attention mechanism to generate the forecasts.</p>
76
+
<p>We propose a novel approach to compute the MAXCUT in attributed graphs, i.e., graphs with features associated with nodes and edges. Our approach is robust to the underlying graph topology and is fully differentiable, making it possible to find solutions that jointly optimize the MAXCUT along with other objectives. Based on the obtained MAXCUT partition, we implement a hierarchical graph pooling layer for Graph Neural Networks, which is sparse, differentiable, and particularly suitable for downstream tasks on heterophilic graphs.</p>
title = {Graph-based Forecasting with Missing Data through Spatiotemporal Downsampling},
104
-
author = {Marisca, Ivan and Alippi, Cesare and Bianchi, Filippo Maria},
105
-
booktitle = {Proceedings of the 41st International Conference on Machine Learning},
106
-
pages = {34846--34865},
107
-
year = {2024},
108
-
volume = {235},
109
-
series = {Proceedings of Machine Learning Research},
110
-
publisher = {PMLR}
98
+
<pre>@article{abate2024maxcutpool,
99
+
title={MaxCutPool: differentiable feature-aware Maxcut for pooling in graph neural networks},
100
+
author={Abate, Carlo and Bianchi, Filippo Maria},
101
+
journal={arXiv preprint arXiv:2409.05100},
102
+
year={2024}
111
103
}
112
104
</pre>
113
105
</div>
@@ -117,35 +109,39 @@ <h3 class="pub-title"><strong>Graph-based Forecasting with Missing Data through
117
109
<divclass="publication">
118
110
119
111
<!-- Title -->
120
-
<h3class="pub-title"><strong>MaxCutPool: differentiable feature-aware Maxcut for pooling in graph neural networks</strong></h3>
112
+
<h3class="pub-title"><strong>Graph-based Forecasting with Missing Data through Spatiotemporal Downsampling</strong></h3>
121
113
122
114
<!-- Authors -->
123
115
<pclass="authors">
124
-
<emstyle="color: gray;">Carlo Abate, Filippo Maria Bianchi</em>
116
+
<emstyle="color: gray;">Ivan Marisca, Cesare Alippi, Filippo Maria Bianchi</em>
125
117
</p>
126
118
127
119
<!-- Venue (optional) -->
128
120
121
+
<pclass="venue">
122
+
<em>ICML 2024</em>
123
+
</p>
124
+
129
125
130
126
<!-- Figure -->
131
127
132
128
<divclass="pub-figure">
133
-
<imgsrc="figs/publications/maxcutpool.png" alt="Figure for MaxCutPool: differentiable feature-aware Maxcut for pooling in graph neural networks">
129
+
<imgsrc="figs/publications/hdtts.png" alt="Figure for Graph-based Forecasting with Missing Data through Spatiotemporal Downsampling">
134
130
</div>
135
131
136
132
137
133
<!-- Abstract -->
138
-
<p>We propose a novel approach to compute the MAXCUT in attributed graphs, i.e., graphs with features associated with nodes and edges. Our approach is robust to the underlying graph topology and is fully differentiable, making it possible to find solutions that jointly optimize the MAXCUT along with other objectives. Based on the obtained MAXCUT partition, we implement a hierarchical graph pooling layer for Graph Neural Networks, which is sparse, differentiable, and particularly suitable for downstream tasks on heterophilic graphs.</p>
134
+
<p>Spatiotemporal graph neural networks achieve striking results by representing the relationships across time series as a graph. Nonetheless, most existing methods rely on the often unrealistic assumption that inputs are always available and fail to capture hidden spatiotemporal dynamics when part of the data is missing. In this work, we tackle this problem through hierarchical spatiotemporal downsampling. The input time series are progressively coarsened over time and space, obtaining a pool of representations that capture heterogeneous temporal and spatial dynamics. Conditioned on observations and missing data patterns, such representations are combined by an interpretable attention mechanism to generate the forecasts.</p>
0 commit comments