More than Meets the Eye: Towards an Artificial Intelligence Observatory
Matias Carrasco Kind, University Illinois Urbana Champaign, United States
Deep Learning for Biometric Systems: Applications and Trends
Fabio Scotti, Computer Science, Universita degli Studi di Milano, Italy
Leveraging Deep Learning for Mental Health Promotion
Elisabeth André, University of Augsburg, Germany
What is Common between Negative Transfer, Uncertainty and Food Recognition?
Petia Radeva, Mathematics and Computer Science, Universitat de Barcelona, Spain
More than Meets the Eye: Towards an Artificial Intelligence Observatory
Matias Carrasco Kind
University Illinois Urbana Champaign
United States
![]()
Brief Bio
Matias Carrasco Kind is currently a Senior Research Scientist at the National Center for Supercomputing Applications (NCSA), Assistant Research Professor in Astronomy and the Associate Director of the Data Science Research Services at the Gies College of Business at the University of Illinois at Urbana-Champaign in the U.S.He is interested in challenging problems involving data intensive science, machine, and deep learning, data mining, data analysis and visualization, image processing, AI generative models, scientific platforms and cyberinfrastructure, data management, software engineering, and scientific cloud computing, among others. Most of his research has been focused on Astrophysics but given the multidisciplinary nature of his work, and the common needs and tools across multiple fields regarding data, he has also applied these techniques to earth sciences, bio-imaging, veterinary, agricultural economics, finance research, and accounting.Matias obtained his PhD in Astronomy with a Computational Science and Engineering option at the University of Illinois which focused on machine learning techniques applied to astronomy at large scales.
Abstract
What if, by leveraging the rapid development of AI, cyber-infrastructure, and astronomical surveys we can create an extremely intelligent machine with infinite knowledge that after being feed with all of the available survey data from all the sources and wavelengths is able to recreate every observation for any object in any wavelength? What if we can feed that entity with an optical image and ask for a radio counterpart? Or ask it to generate infrared data from a given set of properties? Will, that machine been able to make inferences from new observations, assuming its infinite memory?
Even though this might sound too much science-fiction, 10 to 15 years from now might be an incubating project which needs to start today. In this talk, I'll discuss what efforts have been made in this direction, what deep learning advances might help us think in that future, how data from multiple surveys and telescopes can be combined in taking the first steps, and what have we done to make this happen.
Thanks to the advancement of computing, AI, and gateways techniques, the possibilities are countless, and it is now that we need to think about these issues in order to be prepared and to understand how information can be extracted intelligently in favor of scientific discoveries.
Deep Learning for Biometric Systems: Applications and Trends
Fabio Scotti
Computer Science, Universita degli Studi di Milano
Italy
Brief Bio
Fabio Scotti (Senior Member, IEEE) received the Ph.D. degree in computer engineering from the Politecnico di Milano, Milan, Italy, in 2003. He was an Assistant Professor at the Department of Information Technologies, Università degli Studi di Milano, Italy (2002-2015). He was an Associate Professor at the Department of Computer Science, Università degli Studi di Milano, Italy (2015-2020). He is a Full Professor at the Università degli Studi di Milano, Italy since 2020. Original results have been published in over 130 papers in international journals, proceedings of international conferences, books, book chapters, and patents. His current research interests include biometric systems, machine learning and computational intelligence, signal and image processing, theory and applications of neural networks, three-dimensional reconstruction, industrial applications, intelligent measurement systems, and high-level system design.He is an Associate Editor of the IEEE Transactions on Human-Machine Systems and the IEEE Open Journal of Signal Processing. He is serving as Book Editor (Area Editor, section Less-constrained Biometrics) of the Encyclopedia of Cryptography, Security, and Privacy (3rd Edition), Springer. He has been an Associate Editor of the IEEE Transactions on Information Forensics and Security, Soft Computing (Springer) and a Guest Coeditor for the IEEE Transactions on Instrumentation and Measurement.
Abstract
Applications, services and devices using biometric systems are continuously growing, and always new challenges must be faced by researchers. Adaptability, robustness to non-ideal conditions, real-time capability, high accuracy as well as improved interactions with the user, are strong requirements now present in innovative applications and solutions such as cyber security, smart devices and ambient intelligent infrastructures. Biometric systems are designed not just for identity recognition, but they can also be extremely useful to profile users and to understand the human behavior inferring their needs and desires. The presentation will focus on innovative biometric recognition approaches and systems, with specific focus on recent approaches based on artificial intelligence showing how recent deep learning techniques are capable to learn discriminative features directly from complex multidimensional signals and images as well as increasing the accuracy, adaptability, and robustness to non-ideal conditions of biometric systems with respect to traditional approaches. The talk presents biometric systems from a technological point of view and provides an outlook of recent artificial intelligence approaches, including deep learning methods with current strong points and limitations. A review of new applications of biometric systems and recent trends will be also presented.
Leveraging Deep Learning for Mental Health Promotion
Elisabeth André
University of Augsburg
Germany
Brief Bio
Elisabeth André is a full professor of Computer Science and Founding Chair of Human-Centered Artificial Intelligence at Augsburg University in Germany. Elisabeth André has a long track record in multimodal human-machine interaction, embodied conversational agents, social robotics, affective computing and social signal processing. Her work has won many awards including including the Gottfried Wilhelm Leibnitz Prize 2021 of the German Research Foundation, with 2.5 Mio € the highest endowed German research award. In 2010, Elisabeth André was elected a member of the prestigious Academy of Europe, and the German Academy of Sciences Leopoldina. In 2017, she was elected to the CHI Academy, an honorary group of leaders in the field of Human-Computer Interaction. To honor her achievements in bringing Artificial Intelligence techniques to Human-Computer Interaction, she was awarded a EurAI fellowship (European Coordinating Committee for Artificial Intelligence) in 2013. In 2019, she was named one of the 10 most influential figures in the history of AI in Germany by National Society for Informatics (GI). Since 2019, she is serving as the Editor-in-Chief of IEEE Transactions on Affective Computing.
Abstract
An increased number of people face unfavorable mental health conditions at some points in their life, such as depression, stress, and anxiety. The Covid-19 crisis has even reinforced this negative trend. In my talk, I will discuss the potential of modern deep learning techniques to develop a better understanding of mental health conditions. I will describe how to exploit deep learning strategies for developing appropriate intervention strategies, such as personal health care coaches running on the user’s mobile device. Two factors contribute to increased interest in deep learning techniques for mental health promotion. First, the number of affordable mobile and wearable technologies with built-in sensors for quantifying many aspects of our lives is rapidly increasing. Consequently, new data sources and opportunities arise for the development of machine learning models and their applications. Second, deep learning techniques have led to breakthroughs in processing large amounts of data, including multimodal behavioral data and relevant contextual and environmental information. In my talk, I will discuss challenges we need to solve, to implement appropriate solutions for promoting good mental health. Such challenges include the need for lightweight architectures to enable the processing of data on mobile devices, strategies to deal with scarce and noisy data, and the appropriate handling of privacy concerns. Concrete examples from national and international projects, Emma, MindBot, and ForDigitHealth, will illustrate the talk.
What is Common between Negative Transfer, Uncertainty and Food Recognition?
Petia Radeva
Mathematics and Computer Science, Universitat de Barcelona
Spain
![]()
Brief Bio
Prof. Petia Radeva is a Full professor at the Universitat de Barcelona (UB), Head of the Consolidated Research Group “Artificial Intelligence and Biomedical Applications (AIBA)” at the University of Barcelona. Her main interests are in Machine/Deep learning and Computer Vision and their applications to health. Specific topics of interest: data-centric deep learning, uncertainty modeling, self-supervised learning, continual learning, learning with noisy labeling, multi-modal learning, NeRF, food recognition, food ontology, etc. She is an Associate editor in Chief of Pattern Recognition journal and International Journal of Visual Communication and Image Representation. She is a Research Manager of the State Agency of Research (Agencia Estatal de Investigación, AEI) of the Ministry of Science and Innovation of Spain. She supervised 24 PhD students and published more than 100 SCI journal publications and 250 international chapters and proceedings. Petia Radeva belongs to the top 2% of the World ranking of scientists with the major impact in the field of TIC according to the citations indicators of the popular ranking of Stanford. Moreover, she was awarded IAPR Fellow since 2015, ICREA Academia’2015 and ICREA Academia’2022 assigned to the 30 best scientists in Catalonia for her scientific merits, received several international and national awards (“Aurora Pons Porrata”, Prize “Antonio Caparrós” ).
Abstract
Transfer learning can be attributed to several recent breakthroughs in deep learning. It has shown upbeat performance improvements, but most of the transfer learning applications are confined towards fine-tuning. Transfer learning facilitates the learnability of the networks on domains with less data. However, learning becomes a difficult task with complex domains, such as multi-label food recognition, owing to the number of food classes as well as to the fine-grained nature of food images. For this purpose, we will introduce the S2ML-TL, a new transfer learning framework to leverage the knowledge learnt on a simpler single-label food recognition task onto multi-label food recognition. We will show that negative transfer is also related to uncertainty of the classes. After introducing different methods for uncertainty modelling in Deep learning we will show how it can be useful to guide the process of data augmentation and classifier improvement.