Skip to content

Commit 41bccb2

Browse files
dzambondzambon
committed
added tutorial log
1 parent 4f5d4e0 commit 41bccb2

File tree

4 files changed

+185
-1
lines changed

4 files changed

+185
-1
lines changed

‎_data/special_sessions.yml‎

Lines changed: 6 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,11 +1,16 @@
1+
- date: April 23-25, 2025
2+
title: Foundation and Generative Models for Graphs
3+
url: https://www.esann.org/special-sessions#session2
4+
venue: ESANN 2025
5+
city: Bruges, Belgium
16
- date: June 30 - July 5, 2024
27
title: Deep Learning for Graphs
38
url: https://sites.google.com/view/dl4g-2024
49
venue: IEEE WCCI 2024
510
city: Yokohama, Japan
611
- date: October 04-06, 2023
712
title: Graph Representation Learning
8-
url: https://www.esann.org/special-sessions#session4
13+
url: https://www.esann.org/proceedings/2023#488
914
venue: ESANN 2023
1015
city: Bruges, Belgium
1116
- date: July 18-23, 2023

‎_tutorials/graph-based-processing/img/log-logo.png‎

175 KB
Loading

‎_tutorials/graph-based-processing/img/siena-background.jpg‎

610 KB
Loading

‎_tutorials/graph-based-processing/log-2024.html‎

Lines changed: 179 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,179 @@
1+
---
2+
layout: particles_header
3+
title: Graph Deep Learning for Spatiotemporal Time Series
4+
lead: Forecasting, Reconstruction and Analysis
5+
description: The GMLG tutorial on graph deep learning for time-series processing.
6+
venue: LoG Conference 2024
7+
city: Virtual & Siena, Italy
8+
image: tutorials/graph-based-processing/img/thumb.jpg?v=2
9+
---
10+
<!-- Outline section-->
11+
<section>
12+
<div class="container px-4">
13+
<div class="row gx-4 justify-content-center">
14+
<div class="col-lg-8">
15+
<p>
16+
Successful applications of <strong>deep learning</strong> in <strong>time series</strong> processing
17+
often involve training a single neural network on a collection of (related) time series. In the case
18+
of <strong>spatially correlated</strong> time series, pairwise relationships can be modeled by
19+
considering a (possibly dynamic) graph spanning the collection. In this context, <strong>graph-based
20+
methods</strong> take the standard deep learning approach to time series processing a step
21+
forward. The recent theoretical and practical developments in graph neural networks make the
22+
adoption of such an approach particularly appealing and timely.
23+
24+
<!-- Successful applications of <strong>deep learning</strong> in <strong>time series</strong> processing
25+
often involve training a single neural network on a collection of (related) time series. In the case
26+
of <strong>spatially correlated</strong> time series, pairwise relationships can be modeled by
27+
considering a (possibly dynamic) graph spanning the collection. In this context, <strong>graph-based
28+
methods</strong> take the standard deep learning approach to time series processing a step
29+
forward.
30+
31+
Graph representations allow for the conditioning of the predictions w.r.t. neighborhoods of sensors
32+
(i.e., nodes) while learning a single set of shared parameters. The recent theoretical and practical
33+
developments in graph neural networks make the adoption of such an approach particularly appealing
34+
and timely. -->
35+
36+
The twofold <strong>objective</strong> of this tutorial is to:
37+
<ol>
38+
<li>offer a <strong>comprehensive overview</strong> of the field, with a focus on forecasting and
39+
missing data reconstruction;
40+
</li>
41+
<li>provide the necessary theoretically grounded <strong>tools and practical
42+
recommendations</strong> to design and evaluate graph-based spatiotemporal models.
43+
</li>
44+
</ol>
45+
This tutorial is meant for both early-career and experienced researchers looking for a coherent
46+
framing of the field, seeking principles and guidelines to secure the most advanced technologies.
47+
Basic knowledge about deep learning for time series and graph-structured data is sufficient to
48+
benefit from the tutorial.
49+
</p>
50+
</div>
51+
<div class="col-lg-8 text-center">
52+
<h1>Material</h1>
53+
<p>Download the slides used in our tutorial.</p>
54+
<a class="btn btn-primary" href="#" role="button">Download slides (Conf.)</a>
55+
<a class="btn btn-primary" href="#" role="button">Download slides (Meetup)</a>
56+
</div>
57+
</div>
58+
</div>
59+
</section>
60+
<!-- Location section-->
61+
<section class="text-white" style="background-image: url(./img/siena-background.jpg); background-attachment: fixed; background-size: cover;">
62+
<div class="container px-4">
63+
<div class="row gx-4 justify-content-center">
64+
<div class="col-lg-8">
65+
<p class="lead">This tutorial will be delivered mainly at the <strong
66+
class="text-decoration-underline">Learning on Graphs (LoG) Conference</strong>,
67+
held virtually <strong>from the 26th to the 29th of November 2024</strong>, and concludes at the
68+
<strong class="text-decoration-underline">Italy Meetup</strong>, in Siena, Italy,
69+
<strong>from the 4th to the 6th of December 2024</strong>.<br>The tutorial will
70+
take place on
71+
<strong class="text-decoration-underline">Thursday, 28th of November</strong>, 17:00-20:00 (London, UTC+1), and
72+
<strong class="text-decoration-underline">Friday, 6th of December</strong>, 10:30-12:00 (Rome, UTC+2).
73+
</p>
74+
<div class="d-flex flex-column flex-md-row justify-content-between align-items-center"
75+
style="gap: 1.5em">
76+
<a class="btn btn-outline-light" href="https://logconference.org/" role="button"
77+
target="_blank">Conference website</a>
78+
<div>
79+
<img src="./img/log-logo.png" height="182px" class="me-2" />
80+
</div>
81+
<a class="btn btn-outline-light" href="https://sites.google.com/student.unisi.it/log24siena" role="button"
82+
target="_blank">Meetup website</a>
83+
</div>
84+
</div>
85+
</div>
86+
</div>
87+
</section>
88+
<!-- Program section-->
89+
<section class="bg-light">
90+
<div class="container px-4">
91+
<div class="row gx-4 justify-content-center">
92+
<div class="col-lg-8">
93+
<h1 class="text-center">Program</h1>
94+
<h5 class="mt-3 mb-2 fw-light"><span class="fw-bold me-2">Part 1</span> Graph-based processing of
95+
spatiotemporal time series.</h5>
96+
<ol>
97+
<li><strong>Spatiotemporal time series</strong><br>
98+
Definition of the problem settings. Introduction to common downstream tasks: forecasting and
99+
imputation.
100+
</li>
101+
<li><strong>Spatiotemporal graph neural networks (STGNNs)</strong><br>
102+
Presentation of the fundamental components of the general STGNN family of deep learning models
103+
for STS. Recipes and strategies for building effective STGNNs, as well as architectures from the
104+
literature, are provided.
105+
</li>
106+
<li><strong>Global and local models</strong><br>
107+
Discussion on the problem of local effects in spatiotemporal data. Review of the global and
108+
local modeling paradigms with their strengths and practical implications.
109+
</li>
110+
<li><strong>Model quality assessment</strong><br>
111+
Identification of time-space regions, e.g., specific sensors or periods of time, where
112+
predictions can be improved.
113+
</li>
114+
</ol>
115+
116+
<h5 class="mt-3 mb-2 fw-light"><span class="fw-bold me-2">Part 2</span> Challenges and tools.</h5>
117+
<ol>
118+
<li><strong>Latent graph learning</strong><br>
119+
Why and how to learn a graph structure from data when relational information is unavailable,
120+
insufficient or unreliable.
121+
</li>
122+
<li><strong>Learning in non-stationary environments</strong><br>
123+
Challenges and methods associated with modeling the evolution of spatiotemporal systems over
124+
time.
125+
</li>
126+
<li><strong>Scalability</strong><br>
127+
Methods to enable scalability to large sensor networks.
128+
</li>
129+
<li><strong>Dealing with missing data</strong><br>
130+
The problem of missing data in real-world applications and architectures for graph-based
131+
multivariate
132+
time series imputation.
133+
</li>
134+
</ol>
135+
136+
<h5 class="mt-3 mb-2 fw-light"><span class="fw-bold me-2">Demo</span> Coding Spatiotemporal GNNs.</h5>
137+
<p>Overview of open-source Pytorch libraries for graph-based spatiotemporal data processing and short
138+
demo
139+
with Torch Spatiotemporal.</p>
140+
</div>
141+
</div>
142+
</div>
143+
</section>
144+
<!-- Organizers section-->
145+
<section>
146+
<div class="container px-4">
147+
<div class="row gx-4 justify-content-center">
148+
<div class="col-lg-8 text-center">
149+
<h1>Organizers</h1>
150+
<p class="lead">This tutorial is organized by the <a href=”{{site.url}}” class="fw-normal"
151+
target="_blank">GMLG Research
152+
Group</a> within the Swiss AI Lab <a href="https://idsia.ch" class="fw-normal"
153+
target="_blank">IDSIA</a> and <a href="https://usi.ch" class="fw-normal"
154+
target="_blank">Università della Svizzera italiana</a>.</p>
155+
{% assign people = site.data.people %}
156+
<div class="row g-4 my-4 justify-content-center align-items-center text-center text-md-start">
157+
</div>
158+
<div class="row g-4 my-4 justify-content-center align-items-top text-center">
159+
{% assign ca = people | where: "id", "calippi" | first %}
160+
<div class="col-10 col-sm-6 col-md-3 text-center">
161+
<img class="people-thumb medium-thumb" src="{{site.url}}/assets/img/people/{{ca.img}}" />
162+
<h4 class="mt-2">{{ca.name}} <strong>{{ca.surname}}</strong></h4>
163+
<p class="fw-light text-muted mb-1">Professor</p>
164+
{% if ca.links %}
165+
{% include links_list.html links=ca.links %}
166+
{% endif %}
167+
</div>
168+
169+
{% assign dz = people | where: "id", "dzambon" | first %}
170+
{% include people_item.html person=dz hide_description=true col_md=3 %}
171+
{% assign ac = people | where: "id", "acini" | first %}
172+
{% include people_item.html person=ac hide_description=true col_md=3 %}
173+
{% assign im = people | where: "id", "imarisca" | first %}
174+
{% include people_item.html person=im hide_description=true col_md=3 %}
175+
</div>
176+
</div>
177+
</div>
178+
</div>
179+
</section>

0 commit comments

Comments
 (0)