What does graph have to do with machine learning, and data science? A lot, actually, and it goes both ways.
Over the last few years, we have seen what was once a niche research topic ā graph-based machine learning ā snowball. The Year of the Graph was among the first to take stock, point towards this development, and recognize graph-based AI as a key pillar for future development in the field.
In this edition of the YotG Newsletter, we highlight resources focused onĀ graph-based machine learning and data science. Which is not to say thereās lack of news onĀ graph analytics,Ā graph databases, andĀ knowledge graphsĀ ā on the contrary.
2020 has turned Graph Machine Learning into a celebrity of machine learning, argues Michael Bronstein. Arguably, heās a celebrity of Graph Machine Learning himself. A pioneering and prolific researcher at Imperial College London, as well as the Head of Graph Learning Research at Twitter.
Bronstein sought the opinion of prominent researchers in the field of graph ML and its applications trying to summarise the highlights of the past year and predict what is in store for 2021. For a glimpse of Bronsteinās own work, check his Knowledge Connexions presentation āDeep Learning on Graphs: Past, Present, And Futureā
The end of the year is a good time to recap and make predictions. 2020 has turned Graph ML into a celebrity of machine learning. For this post, I sought the opinion of prominent researchers in the field of graph ML and its applications trying to summarise the highlights of the past year and predict what is in store for 2021.
Machine learning can help bootstrap and populate knowledge graphs. The information contained in graphs can boost the efficiency of machine learning approaches. A panel featuring some of the worldās top experts in AI, coming from all sides of the spectrum, got together to discuss howĀ AI + Knowledge are a match made in heaven.
Isabelle Augenstein,Ā Nathan Benaich,Ā Giuseppe Futia,Ā Amy Hodler,Ā Katariina Kari, andĀ Fabio PetroniĀ cover a number of ways graphs, AI and knowledge interact in this 2-hour tour de force.
For theĀ Top Applications of Graph Neural Networks 2021, check our Sergey Ivanovās quick introduction.
What can knowledge-based technologies do for Deep Learning? What is Graph AI, how does it work, what can it do? Whatās next? What are the roadblocks and opportunities?
Whether you are just getting started with graph-based data science and AI, or you are already advanced,Ā Learning with the MachinesĀ is where you can get both inspiration and hands-on knowledge.
Aleksa Gordic will shareĀ How to get started with Graph Machine Learning. Bob van Luijt will explainĀ How businesses apply AI-first solutions in production with the Weaviate Vector Search Engine. Paco Nathan will introduceĀ Graph-Based Data Science, and then deliver aĀ hands-on masterclass.
Tara Safavi will share her work onĀ CoDEx: A Comprehensive Knowledge Graph Completion Benchmark, and Ashleigh Faith will show how toĀ Add more context to machine learning, using taxonomies and knowledge graphs.
Connected Data London Meetup #4
Open source, Graph AI, Search, and Data Science with Python
April 15 ā 16, 2021
When talking about AI, however, be it graph-based or not, thereās semantics involved. And not just in terms of defining what AI is.
AI is notĀ justĀ machine learning. Knowledge-based technologies are also AI. Lately, we started seeing more voices advocating for approaches to bridge the worlds of machine learning and knowledge-based technologies.
Gary Marcus, as he shared in aĀ series of articles on ZDNet, and in hisĀ Knowledge Connexions keynote. Artur dāAvila Garcez and Luis C. Lamb, in their workĀ Neurosymbolic AI: The 3rd Wave, Amit Sheth et. al. InĀ Semantics of the Black-Box: Can Knowledge Graphs help make Deep Learning systems more interpretable and explainable?. And Frank van Harmelen et. al, who createdĀ Modular Design Patterns for Hybrid Learning and Reasoning Systems.
Gary Marcus, a prominent figure in AI, is on a mission to instill a breath of fresh air to a discipline he sees as in danger of stagnating. Knowledge graphs, the 20-year old hype, may have something to offer there.
The last few months have been good for graph databases. A graph database ā Neo4j ā made theĀ Top 20 in DB EnginesĀ for the 1st time. Neo4j also announcedĀ general availability of its Aura managed cloud serviceĀ on GCP, preview on AWS. StardogĀ announced its own cloud DBaaSĀ too.
We had a series of funding rounds, and an upcoming IPO. TigerGraph scoredĀ $105M Series C, Katana GraphĀ $28.5M Series A,Ā Memgraph $6.7MĀ andĀ TerminusDB ā¬3.6M. In the meantime Bitnine, makers of Agens Graph, isĀ working on its IPOĀ ā the first in the market.
Last but not least, we had a round of new releases.Ā AWS open-sourced a Graph NotebookĀ to make working with graph databases easier.Ā Franz Inc released AllegroGraph v7.1,Ā Grakn Labs released Grakn v2.0 Alpha, ,Ā Nebula Graph released v2.0,Ā Ontotext released GraphDB v9.6,Ā RDFox released v.5.0,Ā Stardog released v7.6, andĀ TerminusDB released v4.0.
Everything youāve always wanted to ask about graph databases, but did not have the chance to.
Itās also been a very active period in terms of book publishing. Whether youāre interested in Graph Databases, Data Analytics on Graphs, or Knowledge Graphs and Semantic Technologies, thereās a new book out there for you.
Dave Bechberger and Josh Perryman publishedĀ āGraph Databases in Actionā on Manning. The book introduces graph database concepts by comparing them with relational database constructs. It promises to include just enough theory to get started, then progress to hands-on development.
LjubiÅ”a StankoviÄ et.al.Ā published āData Analytics on Graphsā on Now. The authors revisit graph topologies from a modern data analytics point of view, and proceed to establish a taxonomy of graph networks. With this as a basis, they show how the spectral analysis of graphs leads to even the most challenging machine learning tasks, such as clustering, performed in an intuitive and physically meaningful way.
William L. Hamilton publishedĀ āGraph Representation Learningā on Morgan & Claypool. It aims to serve as If you need a jumping off point to learn more about Graph analytics and the use of graphs in machine learning, as these fields have exploded in the past few years.
Knowledge representation is shorthand for how to represent human symbolic information and knowledge to computers to solve complex questions. KR applications range from semantic technologies and knowledge management and machine learning to information integration, data interoperability, and natural language understanding. Knowledge representation is an essential foundation for knowledge-based AI.
AI capabilities are improving daily, but understanding AIās contextual data and problem-solving approaches is not easy. Graphs are a typically human way of navigating and accessing data.Ā Graphs and AI bring out the best in the symbiosis of humans and systems.
To introduce Knowledge Graphs into Organizations you need to assess the right moment, the organizationās semantic maturity, and to overcome overcome segregation/specialization. Itās important to involve the right stakeholders, andĀ develop KGs in an agile way.
By 2025, graph technologies will be used in 80% of data and analytics innovations, up from 10% in 2021, facilitating rapid decision making
A common use case when working with graphs is converting tabular data to graphs, and adding semantics. From research efforts such as āSemantic Annotation for Tabular Dataā, āInteractively Constructing Knowledge Graphs from Messy User-Generated Spreadsheetsā andĀ DAGOBAH, toĀ Linked Data WizardĀ and implementations of W3CĀ CSV on the WebĀ recommendations, progress is being made.
This enables converting tabular data, and optionally its associated metadata, to a semantic graph in RDF or JSON-LD format. Tabular data includes CSV files, TSV files, and upstream may be coming from spreadsheets, RDBMS export, etc.
Another domain which historically has been prominent for experiments andĀ applications of graphs is SEO. Schema.org had 2 new releases recently,Ā v.11Ā andĀ v.12, summarized by Aaron Bradley and Dan Brickley, respectively.
Search enginesĀ encourage content creators & developers to implement structured data. Structured markup vocabularies include RDFa & SchemaOrg. JSON-LD is more portable, easier to manage, and it hasĀ recently overtaken Microdata.
Knowledge Graphs have huge implications for SEO, content creation and digital marketing across the web. TheĀ Top 5 trends for SEO in 2021Ā include earning your presence in the Knowledge Graph, as it has tremendous impact across multiple platforms.
Ī¤he web as a data graph is a new direction for SEO. Many of the articles that people writing about SEO are about to involve web pages and links between pages. Ī¤his post is about entities and relationships between entities and facts that are written about on web pages, and responses to queries from data graphs on the web about facts and attributes related to entities found on web pages.
Journalists, and data journalists, can use Crowdsourcing, Open data and Knowledge Graphs as a data source, byĀ leveraging Wikidata. A new interconnected ecosystem for research is shaping up.
Science and data are interwoven in many ways. The scientific method has lent a good part of its overall approach and practices to data-driven analytics, software development, and data science. Now data science and software lend some tools to scientific research.
More use cases for graphs, from exotic to everyday. Research claims anĀ AI tool can distinguish between conspiracy theories and true conspiracies, using machine learning and graphs.
Netflix uses Knowledge GraphsĀ for knowledge contained in the content universe. A self-supervised learning task is crafted. Random edges in the graph are selected to form a test set, and the rest of the graph is conditioned to predict missing edges.
A series of related objects are transformed into new graphs via algorithms, generating new abstractions that adhere to a set of transformational rules
Wrapping up with more resources for graph-based machine learning.Ā JraphĀ (pronounced āgiraffeā) is a lightweight library for working with graph neural networks in jax. It provides a data structure for graphs, a set of utilites for working with graphs, and a āzooā of forkable graph neural network models.
Graph Neural Networks explained: Graph neural networks (GNNs) belong to a category of neural networks that operate naturally on data structured as graphs. Despite being what can be a confusing topic, GNNs can be distilled into just a handful of simple concepts.
AutoML systems simplify and democratize AI. Tsinghua University developed an AutoML framework & toolkit designed for graph data and tasks.Ā AutoGLĀ handles all stages of graph learning, may reduce labour & bias in machine learning.
Graph embeddings have become increasingly important in Enterprise Knowledge Graph strategy. What are they, and how are they related to Mowgliās Walk?
To get the Year of the Graph Newsletter in your inbox, subscribeĀ here