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
Copy file name to clipboardExpand all lines: _data/people.yml
+1Lines changed: 1 addition & 0 deletions
Original file line number
Diff line number
Diff line change
@@ -43,6 +43,7 @@ phd:
43
43
- name: "Roberto Neglia"
44
44
image: "/figs/people/rn.jpg"
45
45
bio: "Roberto is a PhD student focusing on basic research in machine learning for time series and graphs. His research interests include randomized architectures for large spatio-temporal models and uncertainty quantification."
Copy file name to clipboardExpand all lines: _data/publications.yml
+71-24Lines changed: 71 additions & 24 deletions
Original file line number
Diff line number
Diff line change
@@ -5,28 +5,27 @@
5
5
arxiv: "https://arxiv.org/abs/2502.09443"
6
6
bibtex: |
7
7
@article{cini2025relational,
8
-
title={Relational Conformal Prediction for Correlated Time Series},
9
-
author={Cini, Andrea and Jenkins, Alexander and Mandic, Danilo and Alippi, Cesare and Bianchi, Filippo Maria},
10
-
journal={arXiv preprint arXiv:2502.09443},
11
-
year={2025}
8
+
title = {Relational Conformal Prediction for Correlated Time Series},
9
+
author = {Cini, Andrea and Jenkins, Alexander and Mandic, Danilo and Alippi, Cesare and Bianchi, Filippo Maria},
10
+
journal = {arXiv preprint arXiv:2502.09443},
11
+
year = {2025}
12
12
}
13
13
14
14
- title: "BN-Pool: a Bayesian Nonparametric Approach to Graph Pooling"
15
15
authors: "Daniele Castellana, Filippo Maria Bianchi"
16
16
figure: "figs/publications/bnpool.gif"
17
-
abstract: "We introduce BN-Pool, the first clustering-based pooling method for GNNs that adaptively determines the number of supernodes in the pooled graph.
18
-
This is done by partitioning the graph nodes into an unbounded number of clusters using a generative model based on a Bayesian non-parametric framework."
17
+
abstract: "We introduce BN-Pool, the first clustering-based pooling method for GNNs that adaptively determines the number of supernodes in the pooled graph. This is done by partitioning the graph nodes into an unbounded number of clusters using a generative model based on a Bayesian non-parametric framework."
- title: "Graph-based Forecasting with Missing Data through Spatiotemporal Downsampling"
@@ -78,3 +78,50 @@
78
78
series = {Proceedings of Machine Learning Research},
79
79
publisher = {PMLR}
80
80
}
81
+
82
+
- title: "The expressive power of pooling in Graph Neural Networks"
83
+
authors: "Filippo Maria Bianchi, Veronica Lachi"
84
+
venue: "NeurIPS 2023"
85
+
figure: "figs/publications/expressive.png"
86
+
abstract: "A graph pooling operator can be expressed as the composition of 3 functions: SEL defines how to form the vertices of the coarsened graph; RED computes the vertex features in the coarsened graph; CON computes the edges in the coarsened graphs. In this work we show that if certain conditions are met on the GNN layers before pooling, on the SEL, and on the RED functions, then enough information is preserved in the coarsened graph. In particular, if two graphs are WL-distinguishable, their coarsened versions will also be WL-dinstinguishable."
title = {The expressive power of pooling in Graph Neural Networks},
92
+
author = {Filippo Maria Bianchi and Veronica Lachi},
93
+
journal = {Advances in neural information processing systems},
94
+
volume = {36},
95
+
year = {2023}
96
+
}
97
+
98
+
- title: "Total Variation Graph Neural Networks"
99
+
authors: "Jonas Berg Hansen, Filippo Maria Bianchi"
100
+
venue: "ICML 2023"
101
+
figure: "figs/publications/tvgnn_larger.png"
102
+
abstract: "We propose the Total Variation GNN model, which can be used to cluster the vertices of an annotated graph, by accounting both for the graph topology and the vertex features. Compared to other GNNs for clustering, TVGNN creates sharp cluster assignments that better approximate the optimal (in the minimum cut sense) partition. The TVGNN model can also be used to implement graph pooling in a deep GNN architecture for tasks such as graph classification."
authors: "Andrea Cini, Ivan Marisca, Filippo Maria Bianchi, Cesare Alippi"
116
+
venue: "AAAI 2023"
117
+
figure: "figs/publications/scalable.png"
118
+
abstract: "SGP is novel approach based on an encode-decoder architecture with a training-free spatiotemporal encoding scheme and where the only learned parameters are in the node-level trainable decoder (an MLP). Representations for each point in time and space can be precomputed and the decoder can be trained by sampling uniformly time and space thus gettig rid of the dependency on sequence lenght and graph size for what concerns the computational complexity of the training procedure. The spatiotemporal encoder relies on two modules: 1) a randomized recurrent neural network for encoding sequences and 2) a propagation process through the graph structure exploiting powers of a graph shift operator."
title: "🌍 Monitoring and Forecasting Geological Activity in Norway"
4
+
keywords: "Spatio-temporal models, time series clustering, earth observation"
5
+
contact_person: "Filippo Maria Bianchi"
6
+
---
7
+
8
+
9
+
## 📝 Description
10
+
This project focuses on analyzing time series of geological activity in Norway, such as landslides, Earth movements, and infrastructure-related displacements, using Earth observation data from InSAR (Interferometric Synthetic Aperture Radar). The time series, available from the [InSAR Norway website](https://insar.ngu.no/), represent ground displacement measured by SAR sensors and are spatially related. The thesis will explore spatio-temporal models to cluster and forecast these time series, aiming to identify areas of high geological activity and detect potential processing errors in the data.
11
+
12
+
The student will work with advanced techniques for time series analysis, including clustering methods to group similar displacement patterns and forecasting models to predict future geological activity. The spatially related nature of the data will be leveraged using spatio-temporal models, which can capture both temporal dynamics and spatial dependencies.
13
+
The student will implement spatiotemporal models using PyTorch and [Torch Spatiotemporal](https://torch-spatiotemporal.readthedocs.io/en/latest/).
14
+
By the end of the project, the student will gain hands-on experience in processing Earth observation data, applying spatio-temporal models, and interpreting results for real-world geological monitoring applications.
15
+
16
+
17
+
**Data:** Time series of ground displacement from InSAR Norway, representing geological activity across various regions in Norway.
This project focuses on the automatic classification of animal behavior using multi-agent trajectory data. The dataset, the [Caltech Mouse Social Interactions](https://sites.google.com/view/computational-behavior/our-datasets/calms21-dataset) (CalMS21), tracks two mice interacting in the same cage, capturing their movements and social behaviors. The task involves classifying their behavior over time, transforming a multivariate time series of trajectories into a sequence of categorical labels (i.e., a seq-to-seq problem with continuous input and categorical output).
11
+
12
+
The student will work with the CalMS21 dataset, which includes trajectory data from videos of freely behaving mice in a resident-intruder assay. The goal is to develop a spatio-temporal model using frameworks like [Torch Spatiotemporal](https://torch-spatiotemporal.readthedocs.io/en/latest/) to automatically classify behaviors, reducing the need for manual labeling by humans. This approach is both innovative and highly relevant, as manual labeling is currently a time-consuming process requiring significant human effort.
13
+
14
+
By the end of the project, the student will gain experience in handling multi-agent trajectory data, designing spatio-temporal models, and applying them to real-world behavior classification tasks. The project has broad implications for behavioral neuroscience and related fields, offering a more efficient and scalable solution for analyzing animal interactions.
15
+
16
+
17
+
**Data:** The Caltech Mouse Social Interactions (CalMS21) Dataset, consisting of trajectory data from videos of interacting mice.
Copy file name to clipboardExpand all lines: _theses/power-flow.md
+1-3Lines changed: 1 addition & 3 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -1,12 +1,10 @@
1
1
---
2
2
layout: base
3
-
title: "Power flow optimization with Graph Neural Networks"
3
+
title: "🔋 Power flow optimization with Graph Neural Networks"
4
4
keywords: "Power flow oprimization, Energy analytics, Power engineering"
5
5
contact_person: "Filippo Maria Bianchi"
6
6
---
7
7
8
-
# Power flow optimization with Graph Neural Networks
9
-
10
8
## 📝 Description
11
9
Power flows are usually optimized with numerical solver that are slow and are not robust to perturbations in the grid topology. GNNs can be used to determine how much power each generator should inject on the grid, based on the grid topology, the physical properties of its components, and the load demand.
The project focuses on applying advanced deep learning techniques to detect leaks in water distribution networks. The data will be represented as spatiotemporal graph, where sensor data is integrated as node and edge features. The student will implement and fine-tune Spatiotemporal Graph Neural Networks (STGNNs) models using PyTorch and [Torch Spatiotemporal](https://torch-spatiotemporal.readthedocs.io/en/latest/), experimenting with different architectures and hyperparameters. By the end of the project, the student will have developed a strong understanding of graph-based modeling, spatiotemporal deep learning, and its application to real-world anomaly detection tasks.
11
+
12
+
13
+
**Data:** An open-source leakage dataset for water distribution networks, which consits of a large number of leakage scenarios, on different water distribution networks, under varying conditions.
0 commit comments