Tutorials
The role of the tutorials is to provide a platform for a more intensive scientific exchange amongst researchers interested in a particular topic and as a meeting point for the community. Tutorials complement the depth-oriented technical sessions by providing participants with broad overviews of emerging fields. A tutorial can be scheduled for 1.5 or 3 hours.
Tutorial on
Deepfakes and Adversarial Machine Learning
Instructor
|
Alessandro Ortis
University of Catania, Department of Mathematics and Computer Science
Italy
|
|
|
|
Brief Bio
Alessandro Ortis is a post-doc researcher at the University of Catania. He has been working in the field of Computer Vision research since 2012, when he joined to the IPLab (Image Processing Laboratory). In 2015 Alessandro was awarded with the Archimede Prize for the excellence of academic career conferred by the University of Catania. In January 2019 he has achieved the PhD in Mathematics and Computer Science, with a PhD Thesis entitled “Methods for Sentiment Analysis and Social Media Popularity of Crowdsourced Visual Contents”. The thesis investigates several aspects related to Visual Sentiment Analysis, applied on crowdsourced images and videos. His PhD has been granted by TIM – Telecom Italia. The main goal of the presented works is to infer the users' preferences by exploiting the crowdsourcing paradigm, as well as to predict the impact of visual contents shared through social media networks. Alessandro spent a part of his PhD as a Visiting Researcher at the Imperial College in London, during which he addressed problems related to the assessment of social influencer advertising campaigns performed through social media. His research interests lie in the fields of Computer Vision, Machine Learning and Multimedia.
|
Abstract
Abstract
In the last few years, deep artificial neural networks have been successfully applied on several problems, especially in computer vision applications. As a consequence, deep learning is being extensively used in most of the recent day-to-day applications. The lecture presents an introduction to Deep Learning methods for vision applications, with a focus on Generative Adversarial Networks and their extensions, which radical advancement recently raised concerns related to the authenticity of multimedia contents and model vulnerability.
The current performances achieved by modern Generative Adversarial Networks put the attention of researcher on two main emerging topics related to multimedia content authenticity (i.e., deepfakes) and vulnerability of applications based on AI technology (i.e., adversarial machine learning). The lecture presents how GANs can be used to generate the so called "deepfakes", which can be used for malicious applications, as well as methods and best practice to detect deepfake contents. The second part of the lecture focuses on the vulnerability of deep neural networks to crafted adversarial examples, which may be imperceptible to the human eye, but can lead the model to misclassify the output. The lecture introduces the main methods for adversarial example generation as well as related countermeasures for model robustness.