Mathematical Sciences Research Seminar Series

Stochastic Separation Theorems: A tool for handling Errors in Artificial Intelligence (AI) Systems

open book on table
Seminars

As part of the School of Science and Technology Mathematical Sciences Research Seminar Series, Professor Ivan Tyukin, University of Leicester presents: Stochastic Separation Theorems: A tool for handling Errors in Artificial Intelligence (AI) Systems

  • From: Wednesday 20 February 2019, 1 pm
  • To: Thursday 21 February 2019, 2 pm
  • Location: 169, Hew Hall Block, Nottingham Trent University, Clifton Campus, Clifton Lane, Nottingham, NG11 8NS

Past event

Event details

As part of the School of Science and Technology Mathematical Sciences Research Seminar Series, Professor Ivan Tyukin, University of Leicester presents: Stochastic Separation Theorems: A tool for handling Errors in Artificial Intelligence (AI) Systems

Abstract

Artificial Intelligence (AI) Systems have now become an everyday reality with industries all over the world offering a rapidly expanding range of AI-based services. Autonomous and self-driving cars, AI assistants for medical diagnostics, and sound and speech recognition are few well-known examples of such services and products.

As an alternative to more traditional approaches whereby a solution or a device is developed on the basis of first physical principles, theory, or laws, many of modern state-of-the art AI systems have been built entirely, or to a significant extent, on the basis of large volumes of empirical data which have been collected, annotated, and stored over time. This empowers such AI systems with unprecedented capabilities to deal with complicated problems by accessing and processing huge quantities of raw information present in the data. At the same time, and due to uncertainties inherent to the data, such AI systems are expected to make an occasional error or mistake. As AI systems spread more widely into our life and society, expected mistakes of AI systems are becoming a part of the new reality.

In this talk we shall consider and review the problem of such mistakes in Artificial Intelligence systems and motivate a mathematical take on this problem. We will present a statement and a solution to the problem of correcting errors in existing AI systems.  Theoretical foundation of the solution is based on Stochastic Separation Theorems and the ideas of measure concentration. We will show that, subject to mild technical assumptions on statistical properties of internal signals in the original AI, with probability close to one, the solution enables instantaneous ''learning away'' of spurious and systematic errors. The solution and the resulting technology will be illustrated with applications to object/face/human shape detection and recognition.

All Welcome

For any enquiries please contact Dr David Chappell.

Location details

Room/Building:

169, Hew Hall Block

Address:

Nottingham Trent University
Clifton Campus
Clifton Lane
Nottingham
NG11 8NS

Past event

Still need help?

+44 (0)115 941 8418