One of the key challenges for Knowledge Graphs is to build, populate and manage data manually—Machine Learning can help with this. Machine Learning analyses the data and algorithms to learn from it, identifying patterns in order to make automated predictions or decisions.
The next obstacle for Knowledge Graphs is processing this at scale as the data continues to grow. With the implementation of Machine Learning and artificial intelligence (AI), Knowledge Graphs can be supplemented with necessary context and structure, in order to make the information more readable and transparent.
The tribes of AI
Facebook has one of the biggest graphs in the world, and is also one of the leaders in Machine Learning and AI. Sebastian Riedel, Researcher at Facebook AI Research, helped explore how knowledge-based approaches can connect with, and contribute to Machine Learning for better AI.
Participating in a panel of experts exploring AI tribes and interconnections, and the role of graphs, the Facebook AI researcher deliberated at the Connected Data London event that:
“There are several tribes that I like; there’s the Connectionist Tribe, Analogue Reasoning Tribe, Symbolic Reasoning Tribe, Evolutionary Tribe, and Bayesian Tribe. I’d say that I fit into the Connectionist, Bayesian and Symbolic Tribe.
“I’d start with the Connectionist Tribe, this is about Deep Learning which is this idea of building new networks which essentially do everything for you—if you give them enough data.
In the last couple of years, we’ve seen amazing results coming out of that line of work, in particular, whenever there’s a lot of training data for the particular task you’re looking at.
“I think there is this big problem of marrying these different worlds of five tribes; my focus is on two: Connectionist and Symbolic. There is a big community within ML [Machine Learning] that just looks at this problem, and has been looking for the last twenty, thirty years how Machine Learning and Symbolic AI can come together.”
Neural reasoning and Knowledge Graphs
Octavian is one of the pioneers in new approaches to Machine Reasoning and Graph-based Learning. They are working to build machines that can answer useful questions, using neural reasoning and Knowledge Graphs. Andy Jefferson, Neural Networks and Graph AI Researcher at Octavian, provides an insight into the state of the art in building AI with Graphs.
“Machine Learning has a pretty simple algorithm at the top level; we have some training, a model which is our concept of the world and these are the neural networks that we are talking about—then we have data.
“To start off with the model has no priors, we don’t program the model to have an understanding of grammar. If we’ re working with text, we don’t program the model to have open CV style concepts of edges and spheres with images.”
The early days of a semantic web
The Open Data Institute (ODI) is one of the leaders in using data in the public sector. Sir Nigel Shadbolt from the ODI has been working on AI for many decades. Now that AI is in the limelight, let’s not forget all the long years of foundational research and the effort that went into accumulating data that facilitates AI. In his keynote, Sir Nigel Shadbolt emphasized the importance of Linked Data as critical infrastructure.
“In 2003, we imagined a Semantic web which did not yet exist, in which we will be able to reintegrate heterogeneous information from a wide variety of sources and almost treated the web as a distributed knowledge space.
“People weren’t using Linked Data back in the day, so we harvested content from around the websites of all the departments of computer science in the country, within this thing called CS active space—an RDF-based, graph-based representation of the world of computing.
“In 2005, we began our first work with the public sector data-set, we took data from lots of London authorities and services, and began to triplify them, put them in graph stores. That work was represented to Parliament in 2007.
“To this day, we have police.uk, which is all the reported crime data for any particular zone and place—and this was sensitive data that we’d pinpoint where exactly issues were occurring.”
AI is massively hyped these days, and not without reason, as progress and achievements have been outstanding. But let us not forget that there is no single way or method to achieve results: AI is not all about Machine Learning. Context, structure, and reasoning are necessary ingredients, and Knowledge Graphs and Linked Data are key technologies for this.