We are delighted to invite you to a talk on GAN, Deepfake and Misinformation, with perspectives from both academia and industry and an opportunity to join in the conversation over the selected topic.

Over 90mn, the following speakers will discuss their work and take questions:

• Elena Kochkina, Postdoctoral Research Assistant in EECS, will talk about State of the art and challenges of automated rumour verification in social media conversations;

• Steven Allen, Senior security consultant & Cloud Platform security manager at Capgemini, will answer the question “What can AI learn from Biometrics?”;

• Ioannis Patras, Professor in Computer Vision and Human Sensing in EECS will discuss his research on the GAN technology, detecting near duplicate videos and associated issues;

• Chris Reed, Professor of Electronic Commerce Law, Centre for Commercial Law Studies, at QMUL will talk about Misleading humans via AI.

The event will be introduced by Prof Steve Uhlig, Head of School, and chaired by Prof Sean Gong, chair of the EECS Industrial Board.

The purpose of this event is to offer a space for knowledge sharing between the EECS research community, our industry partners, and guests.

A short abstract of each presentation is available below.

Zoom access details here below. Join Zoom Meetinghttps://qmul-ac-uk.zoom.us/j/88654686151?pwd=KzlOblZJRVk3Z3VIdSs0dWJ1bkJ5dz09

Meeting ID: 886 5468 6151Passcode: 647691

State of the art and challenges of automated rumour verification in social media conversations, Elena Kochkina, Postdoctoral Research Assistant, EECS QMUL.

Abstract: Online misinformation had a dramatic impact on our society in recent years, affecting people in many sectors, from finance, health to politics. The importance of combatting the spread of fake news has been acknowledged by the public, companies and governments. The problem is being addressed through policies, journalistic work, media literacy of the population, social media platforms response and research. Our research aims to tackle this problem using Machine Learning and Natural Language Processing methods to propose systems for automated rumour verification that would assist journalists in their work. In this talk, we will discuss various types of existing approaches to automated rumour verification, overview recent advances and outline open challenges that rumour verification models are facing, and share views on how to tackle them.

Title: What can AI learn from Biometrics? Steven Allen, Senior security consultant & Cloud Platform security manager at Capgemini.

Abstract: While biometrics have become ubiquitous now, they have taken years to get there and still have not maximised their potential. AI shares some features with biometrics – will AI vendors and organisations using AI learn from the biometric experience or will they make the same mistakes?

Title: GAN the technology, detecting near duplicate videos and associated issues, Ioannis Patras, Professor in Computer Vision and Human Sensing, EECS, QMUL.

Abstract: Recent years have witnessed an unprecedented interest in the generation and analysis of visual content, be that images or video. In this talk, we will present some recent advances in this area using Machine Learning and Computer Vision methods. In particular, I will talk about recent advances in generation of images and on unsupervised ways of controlling the generation process. I will present some work on learning to compare image sequences and how these are applied for detection of near duplicates videos, and on action recognition.

Title: Misleading humans via AI, Chris Reed, Professor of Electronic Commerce Law, Centre for Commercial Law Studies, QMUL.

Abstract: Deepfakes produce convincing images, audio and video, which never happened. AI bots generate false information and disseminate it widely. Targeting algorithms push fake and carefully selected ‘news’ to susceptible groups with high granularity. All these trends are worrying lawmakers greatly. There is a growing move (such as the UK Online Harms initiative) to place legal responsibilities on platforms to detect these kinds of misinformation, and either take them down rapidly or even prevent them being communicated in the first place. Failure to do so, which might include a responsibility to develop AI technologies to help, will lead to substantial penalties.

Because platforms will tend to be over-cautious, they are likely to put pressure on AI developers to make the products of their technologies easily detectable. Platform users will also be pressured to identify and limit their use of AI. All this will be done via terms and conditions, rather than published law, making the platforms into new kinds of lawmaker.

GAN, Deepfake and misinformation