Network Repository
A cool interactive repository of about a thousand of different graphs. Could be useful if you need some graphs with specific properties for specific tasks.
A cool interactive repository of about a thousand of different graphs. Could be useful if you need some graphs with specific properties for specific tasks.
Network Data Repository
Network Data Repository | The First Interactive Network Data Repository
The first interactive network dataset repository with interactive graph visualization and analytics
Fresh picks from ArXiv
Today at ArXiv smoothing for link prediction, design space of GNNs, and scalable networks for heterogeneous graphs ⛰
If I forgot to mention your paper, please shoot me a message and I will update the post.
Conferences
- Node Similarity Preserving Graph Convolutional Networks WSDM 2021
- Design Space for Graph Neural Networks with Jure Leskovec, NeurIPS 2020
Graphs
- VLG-Net: Video-Language Graph Matching Network for Video Grounding
- Scalable Graph Neural Networks for Heterogeneous Graphs
- Graph embeddings via matrix factorization for link prediction: smoothing or truncating negatives?
- Reinforcement Learning of Graph Neural Networks for Service Function Chaining
- Quantum algorithms for learning graphs and beyond
Survey
- Survey and Open Problems in Privacy Preserving Knowledge Graph: Merging, Query, Representation, Completion and Applications
- Subpath Queries on Compressed Graphs: a Survey
Today at ArXiv smoothing for link prediction, design space of GNNs, and scalable networks for heterogeneous graphs ⛰
If I forgot to mention your paper, please shoot me a message and I will update the post.
Conferences
- Node Similarity Preserving Graph Convolutional Networks WSDM 2021
- Design Space for Graph Neural Networks with Jure Leskovec, NeurIPS 2020
Graphs
- VLG-Net: Video-Language Graph Matching Network for Video Grounding
- Scalable Graph Neural Networks for Heterogeneous Graphs
- Graph embeddings via matrix factorization for link prediction: smoothing or truncating negatives?
- Reinforcement Learning of Graph Neural Networks for Service Function Chaining
- Quantum algorithms for learning graphs and beyond
Survey
- Survey and Open Problems in Privacy Preserving Knowledge Graph: Merging, Query, Representation, Completion and Applications
- Subpath Queries on Compressed Graphs: a Survey
How node features affect performance of GNN?
This is an open question that I recently thought a bit. In particular, what surprised me are the results from a recent paper on Label Propagation on a particular dataset Rice31 (table below).
You can see that some models achieve 80% accuracy, while others 10% (random guess). In the paper they say that the node features are heterogeneous features such as gender or major, but after speaking with authors it seems they use spectral embeddings instead.
I have tried this dataset with GNN and my results are close to random guess (10%). I tried several variations of GNN as well as node features, but didn't get much higher than 15%. Then I tried GBDT with spectral embeddings and it gave me about 50% accuracy. I haven't tried LP yet on this dataset, but it would be remarkable to see that LP with spectral embeddings can have such a drastic difference with GNN.
This and other experiments led me to think that the paradigm of message passing is too strong, i.e. aggregating information simultaneously among your neighbors may not be a good idea in general. The inductive bias that such model has could be wrong for a particular graph dataset. GNN work on some graph datasets, but how node labels depend on the graph structure is very similar to how message-passing works. In other words, if you were to create a dataset, where a node label equals to an average label of your neighbors, then GNN that does average aggregation would easily learn such dependency. But if your node labels depend on the structure in some counter-intuitive way (for example, by picking a neighbor at random and then assigning its node label), then your GNN with average aggregation would fail. In other words, GNN models don't have to follow message-passing paradigm, they can have very different design principles and that's something that I think we will see in the coming years.
This is an open question that I recently thought a bit. In particular, what surprised me are the results from a recent paper on Label Propagation on a particular dataset Rice31 (table below).
You can see that some models achieve 80% accuracy, while others 10% (random guess). In the paper they say that the node features are heterogeneous features such as gender or major, but after speaking with authors it seems they use spectral embeddings instead.
I have tried this dataset with GNN and my results are close to random guess (10%). I tried several variations of GNN as well as node features, but didn't get much higher than 15%. Then I tried GBDT with spectral embeddings and it gave me about 50% accuracy. I haven't tried LP yet on this dataset, but it would be remarkable to see that LP with spectral embeddings can have such a drastic difference with GNN.
This and other experiments led me to think that the paradigm of message passing is too strong, i.e. aggregating information simultaneously among your neighbors may not be a good idea in general. The inductive bias that such model has could be wrong for a particular graph dataset. GNN work on some graph datasets, but how node labels depend on the graph structure is very similar to how message-passing works. In other words, if you were to create a dataset, where a node label equals to an average label of your neighbors, then GNN that does average aggregation would easily learn such dependency. But if your node labels depend on the structure in some counter-intuitive way (for example, by picking a neighbor at random and then assigning its node label), then your GNN with average aggregation would fail. In other words, GNN models don't have to follow message-passing paradigm, they can have very different design principles and that's something that I think we will see in the coming years.
Telegram
Graph Machine Learning
Combining Label Propagation and Simple Models Out-performs Graph Neural Networks
This paper by Cornell and Facebook made a lot of noise on Twitter recently. In short, it shows that GNNs can be outperformed by simpler models such as MLP + Label Propagation…
This paper by Cornell and Facebook made a lot of noise on Twitter recently. In short, it shows that GNNs can be outperformed by simpler models such as MLP + Label Propagation…
Jraph - A library for graph neural networks in jax.
Jgraph is a new library by DeepMind for constructing GNNs in JAX (autograd computation) and Haiku (writing neural network layers). Could be useful if you cannot use PyTorch.
Jgraph is a new library by DeepMind for constructing GNNs in JAX (autograd computation) and Haiku (writing neural network layers). Could be useful if you cannot use PyTorch.
GitHub
GitHub - google-deepmind/jraph: A Graph Neural Network Library in Jax
A Graph Neural Network Library in Jax. Contribute to google-deepmind/jraph development by creating an account on GitHub.
Graph Machine Learning research groups: Jimeng Sun
I do a series of posts on the groups in graph research, previous post is here. The 19th is Jimeng Sun, the head of SunLab at UIUC, teaching the courses of Big Data Analytics and Healthcare as well as Computing and Society.
Jimeng Sun (~1981)
- Affiliation: University of Illinois Urbana-Champaign
- Education: Ph.D. at CMU in 2002 (advisor: Christos Faloutsos)
- h-index 66
- Awards: KDD, ICDM, SDM best paper awards
- Interests: drug discovery, GNNs, graph mining
I do a series of posts on the groups in graph research, previous post is here. The 19th is Jimeng Sun, the head of SunLab at UIUC, teaching the courses of Big Data Analytics and Healthcare as well as Computing and Society.
Jimeng Sun (~1981)
- Affiliation: University of Illinois Urbana-Champaign
- Education: Ph.D. at CMU in 2002 (advisor: Christos Faloutsos)
- h-index 66
- Awards: KDD, ICDM, SDM best paper awards
- Interests: drug discovery, GNNs, graph mining
Telegram
Graph Machine Learning
Graph Machine Learning research groups: Albert-László Barabási
I do a series of posts on the groups in graph research, previous post is here. The 18th is Albert-László Barabási, who introduced scale-free networks and in particular Barabási–Albert model and…
I do a series of posts on the groups in graph research, previous post is here. The 18th is Albert-László Barabási, who introduced scale-free networks and in particular Barabási–Albert model and…
Planarity game
If you need some time to procrastinate and you want to do it with graphs, here is a fun game to play, called Tronix2. You just need to make the drawn graphs planar. There are several clones of this game (here and here), which even explain how to generate planar graphs. And here is Numberphile video about planar graphs.
If you need some time to procrastinate and you want to do it with graphs, here is a fun game to play, called Tronix2. You just need to make the drawn graphs planar. There are several clones of this game (here and here), which even explain how to generate planar graphs. And here is Numberphile video about planar graphs.
Atom
Treksit.com is for sale at Atom!
Domain name Treksit.com evokes a sense of adventure and exploration, making it ideal for travel, outdoor, or tech startups. The word trek suggests a journey or expedition, while sit implies a place to relax and reflect. This duality can app . " #CatchyDomains…
Golden Knowledge Graph
Golden is a Silicon Valley startup building a knowledge database (similar to Wikipedia) — a good example how knowledge graphs can be commercialized.
Golden is a Silicon Valley startup building a knowledge database (similar to Wikipedia) — a good example how knowledge graphs can be commercialized.
Fresh picks from ArXiv
Today at ArXiv: fast algorithms for dynamic graphs, new adversarial attacks, and GNNs for generating dances 💃
If I forgot to mention your paper, please shoot me a message and I will update the post.
GNN
- Curvature Regularization to Prevent Distortion in Graph Embedding NeurIPS 2020
- Fast and Uncertainty-Aware Directional Message Passing for Non-Equilibrium Molecules with Stephan Günnemann, Workshop NeurIPS 2020
- A Targeted Universal Attack on Graph Convolutional Network
- Fast and Accurate Anomaly Detection in Dynamic Graphs with a Two-Pronged Approach with Christos Faloutsos
- Near-Optimal Algorithms for Reachability, Strongly-Connected Components and Shortest Paths in Partially Dynamic Digraphs
Applications
- Video Self-Stitching Graph Network for Temporal Action Localization
- Learning to dance: A graph convolutional adversarial network to generate realistic dance motions from audio
- RetroGNN: Approximating Retrosynthesis by Graph Neural Networks for De Novo Drug Design Workshop NeurIPS 2020
- Quantifying Explainers of Graph Neural Networks in Computational Pathology
Today at ArXiv: fast algorithms for dynamic graphs, new adversarial attacks, and GNNs for generating dances 💃
If I forgot to mention your paper, please shoot me a message and I will update the post.
GNN
- Curvature Regularization to Prevent Distortion in Graph Embedding NeurIPS 2020
- Fast and Uncertainty-Aware Directional Message Passing for Non-Equilibrium Molecules with Stephan Günnemann, Workshop NeurIPS 2020
- A Targeted Universal Attack on Graph Convolutional Network
- Fast and Accurate Anomaly Detection in Dynamic Graphs with a Two-Pronged Approach with Christos Faloutsos
- Near-Optimal Algorithms for Reachability, Strongly-Connected Components and Shortest Paths in Partially Dynamic Digraphs
Applications
- Video Self-Stitching Graph Network for Temporal Action Localization
- Learning to dance: A graph convolutional adversarial network to generate realistic dance motions from audio
- RetroGNN: Approximating Retrosynthesis by Graph Neural Networks for De Novo Drug Design Workshop NeurIPS 2020
- Quantifying Explainers of Graph Neural Networks in Computational Pathology
Undergraduate Math Student Pushes Frontier of Graph Theory
A new article at QuantaMagazine about 21 year old who improved results of Erdős and Szekeres on the upper bound for two-color Ramsey numbers. Informally, Ramsey numbers can be explained as "how big graphs can get before patterns inevitably emerge". This is in addition to the recent proof for lower bounds, also covered in Quanta.
A new article at QuantaMagazine about 21 year old who improved results of Erdős and Szekeres on the upper bound for two-color Ramsey numbers. Informally, Ramsey numbers can be explained as "how big graphs can get before patterns inevitably emerge". This is in addition to the recent proof for lower bounds, also covered in Quanta.
Quanta Magazine
Undergraduate Math Student Pushes Frontier of Graph Theory
At 21, Ashwin Sah has produced a body of work that senior mathematicians say is nearly unprecedented for a college student.
ICLR 2021 Graph Papers
Here is a list of graph papers with their final scores. This is in addition to the list for all the papers. Overall, 74 papers (out of 208 graph papers) increased their scores and 18 decreased.
Here is a list of graph papers with their final scores. This is in addition to the list for all the papers. Overall, 74 papers (out of 208 graph papers) increased their scores and 18 decreased.
Twitter
Sergey Ivanov
#ICLR2021 reviews are finalized and below is a final table with scores. Here is a teaser of how rebuttal helps the papers. 🧵1/n https://t.co/GISD8ukXWs
Geometric Deep Learning: Successes, Challenges, Next Steps
A good video overview of the field by Michael Bronstein: talking about deriving convolution from the first principles, first GNN models, expressivity of GNNs, and the future applications of the field.
A good video overview of the field by Michael Bronstein: talking about deriving convolution from the first principles, first GNN models, expressivity of GNNs, and the future applications of the field.
YouTube
DVL Guest Lectures - Michael Bronstein, Imperial College London
Guest lecture by Michael Bronstein (Imperial College London, USI, Twitter)
Abstract: Geometric deep learning has recently become one of the hottest topics in machine learning, with its particular instance, graph neural networks, being used in a broad spectrum…
Abstract: Geometric deep learning has recently become one of the hottest topics in machine learning, with its particular instance, graph neural networks, being used in a broad spectrum…
Open Access Theses and Dissertations
Seeking for an inspiration for your dissertation or maybe want to check the latest monolithic works in graph community, take a look at OATD portal. Here is for example a search for all dissertations that have
Seeking for an inspiration for your dissertation or maybe want to check the latest monolithic works in graph community, take a look at OATD portal. Here is for example a search for all dissertations that have
graph in their title, resulting in ~400 PhD and ~100 MSc theses just in 2016-2020 period.Don't crack under pressure
I remember when I was interning at HKUST during my master's years I had a chance to see a motivational presentation for freshmen from one of the tenured professors there. One of the things he was emphasizing is that PhD is stressful experience with lots of uncertainty and you should keep being focused, brave and don't crack under the pressure. Here is a funny story that demonstrates it. A guy realized he has a bug 2 weeks before submitting his PhD thesis and asked MathOverflow community to help him to fix it, luckily it worked out well.
I remember when I was interning at HKUST during my master's years I had a chance to see a motivational presentation for freshmen from one of the tenured professors there. One of the things he was emphasizing is that PhD is stressful experience with lots of uncertainty and you should keep being focused, brave and don't crack under the pressure. Here is a funny story that demonstrates it. A guy realized he has a bug 2 weeks before submitting his PhD thesis and asked MathOverflow community to help him to fix it, luckily it worked out well.
MathOverflow
Issue UPDATE: in graph theory, different definitions of edge crossing numbers - impact on applications?
QUICK FINAL UPDATE: Just wanted to thank you MO users for all your support. Special thanks for the fast answers, I've accepted first one, appreciated the clarity it gave me. I've updated my torus
Graph Mining & Learning Workshop NeurIPS 2020
NeurIPS 2020 just started and there is a good workshop by Google Research on graph mining (by Bryan Perozzi and others). To see the videos you need to register for expo but it's free. Then you will have links to videos. There is also a slide deck with 312 pages about many interesting topics.
Update: it seems there is a bug in the registration panel, so you can access the schedule and videos in this page.
NeurIPS 2020 just started and there is a good workshop by Google Research on graph mining (by Bryan Perozzi and others). To see the videos you need to register for expo but it's free. Then you will have links to videos. There is also a slide deck with 312 pages about many interesting topics.
Update: it seems there is a bug in the registration panel, so you can access the schedule and videos in this page.
Graph Mining & Learning
Graph Mining and Learning @ NeurIPS
at the 2020 Neural Information Processing Systems Conference
Fresh picks from ArXiv
Today at ArXiv: approximate descent for training GNNs, expressive power of GNN, and GNN-doctor 💉
If I forgot to mention your paper, please shoot me a message and I will update the post.
GNN
- Learning Graph Neural Networks with Approximate Gradient Descent AAAI 2021
- Attention-Driven Dynamic Graph Convolutional Network for Multi-Label Image Recognition ECCV 20
- Learning Node Representations from Noisy Graph Structures ICDM 2020
- Accelerated Charged Particle Tracking with Graph Neural Networks on FPGAs Workshop NeurIPS 2020
- Graph Neural Networks for Improved El Niño Forecasting Workshop NeurIPS 2020
Applications
- Bayesian Graph Neural Networks for Molecular Property Prediction
- Dr-COVID: Graph Neural Networks for SARS-CoV-2 Drug Repurposing
- Partial Gromov-Wasserstein Learning for Partial Graph Matching
- Counting Substructures with Higher-Order Graph Neural Networks: Possibility and Impossibility Results with Stefanie Jegelka
- Graph Mixture Density Networks
Survey
- A Survey on Heterogeneous Graph Embedding: Methods, Techniques, Applications and Sources
Today at ArXiv: approximate descent for training GNNs, expressive power of GNN, and GNN-doctor 💉
If I forgot to mention your paper, please shoot me a message and I will update the post.
GNN
- Learning Graph Neural Networks with Approximate Gradient Descent AAAI 2021
- Attention-Driven Dynamic Graph Convolutional Network for Multi-Label Image Recognition ECCV 20
- Learning Node Representations from Noisy Graph Structures ICDM 2020
- Accelerated Charged Particle Tracking with Graph Neural Networks on FPGAs Workshop NeurIPS 2020
- Graph Neural Networks for Improved El Niño Forecasting Workshop NeurIPS 2020
Applications
- Bayesian Graph Neural Networks for Molecular Property Prediction
- Dr-COVID: Graph Neural Networks for SARS-CoV-2 Drug Repurposing
- Partial Gromov-Wasserstein Learning for Partial Graph Matching
- Counting Substructures with Higher-Order Graph Neural Networks: Possibility and Impossibility Results with Stefanie Jegelka
- Graph Mixture Density Networks
Survey
- A Survey on Heterogeneous Graph Embedding: Methods, Techniques, Applications and Sources
MoleculeKit: Machine Learning Methods for Molecular Property Prediction and Drug Discovery
MoleculeKit is a new framework that deals with molecule predictions. It represents molecules as both graphs and sequences and then apply GNN or kernel together with BERT for downstream molecular tasks (predicting properties of nodes or graphs).
MoleculeKit is a new framework that deals with molecule predictions. It represents molecules as both graphs and sequences and then apply GNN or kernel together with BERT for downstream molecular tasks (predicting properties of nodes or graphs).
Privacy-Preserving Deep Learning Over Graphs
60 slides of overview of the emerging field of privacy-preserving GNNs. Could be interesting if you search for a new research topic.
60 slides of overview of the emerging field of privacy-preserving GNNs. Could be interesting if you search for a new research topic.
Graph Machine Learning research groups: Max Welling
I do a series of posts on the groups in graph research, previous post is here. The 20th is Max Welling, the head of the Amsterdam Machine Learning Lab. He co-founded a startup Scyfer BV that was acquired by Qualcomm, where he serves as VP of technologies. Max has a diverse research interests, including lately developments in graph machine learning field.
Max Welling (1968)
- Affiliation: University of Amsterdam, Qualcomm
- Education: Ph.D. at Utrecht University in 1998 (advisor: Gerard 't Hooft)
- h-index 73
- Awards: ECCV Koenderink Prize, ICML best papers.
- Interests: equivariant networks, variational encoders, GNNs.
I do a series of posts on the groups in graph research, previous post is here. The 20th is Max Welling, the head of the Amsterdam Machine Learning Lab. He co-founded a startup Scyfer BV that was acquired by Qualcomm, where he serves as VP of technologies. Max has a diverse research interests, including lately developments in graph machine learning field.
Max Welling (1968)
- Affiliation: University of Amsterdam, Qualcomm
- Education: Ph.D. at Utrecht University in 1998 (advisor: Gerard 't Hooft)
- h-index 73
- Awards: ECCV Koenderink Prize, ICML best papers.
- Interests: equivariant networks, variational encoders, GNNs.
Telegram
Graph Machine Learning
Graph Machine Learning research groups: Jimeng Sun
I do a series of posts on the groups in graph research, previous post is here. The 19th is Jimeng Sun, the head of SunLab at UIUC, teaching the courses of Big Data Analytics and Healthcare as well as Computing…
I do a series of posts on the groups in graph research, previous post is here. The 19th is Jimeng Sun, the head of SunLab at UIUC, teaching the courses of Big Data Analytics and Healthcare as well as Computing…
Deep Graph Networks Reading Group
There is a reading group at Bicocca University (Milan, Italy). Next session will happen on Monday, 14th December at 10am (UK time). The paper "HATS a hierarchical graph attention network for stock movement prediction" will be discussed. If you want to join you can get a link by contacting @Sagax_ita or via [email protected].
There is a reading group at Bicocca University (Milan, Italy). Next session will happen on Monday, 14th December at 10am (UK time). The paper "HATS a hierarchical graph attention network for stock movement prediction" will be discussed. If you want to join you can get a link by contacting @Sagax_ita or via [email protected].
Google
Dimitri Ognibene's Homepage - Graph Net - Reading Group
Deep Graph Networks Reading Group