Skip to main content

Samsung Research UK and the University of Oxford – Collaborating on Knowledge Representation to build User-Oriented AI

Posted:

Samsung Research UK (SRUK) have announced a collaboration with Professor Ian Horrocks of the University of Oxford to develop knowledge representation techniques to advance the next generation of AI solutions, making Samsung products and services more helpful and personalized for users.

While neural-network-based ML techniques are making great advances in applications such as speech recognition and image understanding, a fundamental gap remains in developing AI services that can use facts and reasoning to better understand the world around us. As an example, a kitchen AI assistant may need to understand that while a recipe might call for a '400g tin of tomatoes' as an ingredient, it is necessary to open the tin and pour out the tomatoes inside rather than placing the entire unopened tin of tomatoes into a saucepan. Such common-sense reasoning about the world is completely unconscious and natural for human beings, but has historically been a real impediment to AI services that aim to be of use in the world.

A key component of this challenge is the use of logical reasoning. If a kitchen AI system has an understanding that containers are a distinct class of object from the foods they contain, then resolving the ambiguity of a '400g tin of tomatoes' becomes easier. Such a collection of facts about the world, together with rules that specify the relationships between classes of objects, is called an ontology. Ontologies have been studied for many years by the logical-reasoning AI community and a large body of work exists that studies the use of formal reasoning algorithms on structured ontologies to derive new facts and understanding.

Ontologies are heavily used, for example, in the medical domain, to specify diseases, anatomic parts, medical procedures, drugs and other entities that need to be specified precisely and consistently across different applications. In a more everyday setting, an ontology could be used to represent knowledge about different chronic diseases, allergies or diets that a kitchen assistant might need to be aware of in order to usefully recommend diet plans or recipes to users. If a user suffers from hypertension, the AI system could use knowledge about chronic diseases and diet to understand that this user should limit their salt intake, and that as well as avoiding foods that directly contain high amounts of salt, such as salted peanuts, the user should avoid recipes that contain ingredients such as soy sauce.

SRUK’s collaboration with the University of Oxford aims to develop fundamental technologies to incorporate knowledge representation via ontologies into the next generation of Samsung AI products and services. SRUK will be working with world-renowned ontology expert, Professor Ian Horrocks and his group at the Department of Computer Science at the University of Oxford. The collaboration is part of Samsung’s advanced research funding programme that looks to develop the next generation of technologies and incorporate them into Samsung products.

Professor Horrocks’ research in the area of knowledge representation and logical reasoning spans over three decades. His work on reasoning in logical frameworks forms the basis of many powerful reasoning tools available today. He played a central role in the development of the Web Ontology Language and has served as program chair for the ISWC conference, one of the most highly-regarded forums for AI research. Together with his students and fellow researchers at the Department of Computer Science at the University of Oxford he continues to publish ground-breaking research in the field of logical AI.

One of the key challenges this collaboration will address is integration of logical AI and statistical machine learning. Professor Horrocks’ team at Oxford have been researching and developing advanced techniques to integrate knowledge from different sources and produce a broader and more powerful ontology representation. SRUK will use its experience with machine learning natural language processing (NLP) to improve the current state-of-the-art in this field. NLP techniques could be used to better understand that while the strings 'salt' and 'sodium chloride' are not alike, they occur interchangeably in text, and that they likely refer to the same substance in the real world. This semantic matching could then be used to better integrate knowledge bases that might use inconsistent labels for the same ingredient.

By integrating techniques from the logical-reasoning AI field and the machine-learning NLP field, SRUK and Oxford’s collaboration looks to bring the best from both worlds to create powerful knowledge representation that will drive the next generation of Samsung products and services.

  

By Eoghan Flanagan / Samsung Research UK