Paulo Quaresma, University of Évora

AI and NLP: (good) challenges, (some) threats and (many) opportunities

Monday, September 25th, 14.30

Abstract: In this presentation, some of the main challenges, threats and opportunities in the area of Natural Language Processing will be discussed, namely with the creation and use of LLM (Large Language Models). A special focus will be placed on the (many) capabilities and the (also existing) limitations of systems based solely on LLMs, highlighting research opportunities and open challenges. In addition, some of the developed and ongoing projects in the area of NLP in the research group "VISTA Lab -- Video, Image, Speech, and Text Analysis Lab" at the University of Évora will be presented.

Wolfgang Banzhaf, Michigan State University

Is Evolutionary Machine Learning the next frontier in AI?

Tuesday, September 26th, 9.00

Abstract: Deep neural networks as a bio-inspired computing method have been incredibly successful over the last two decades, to the point where AI and deep learning or often used interchangeably. In this talk I shall discuss another basic paradigm derived from biology, Natural Evolution, as it is applied in computation and its advantages for Artificial Intelligence.  This field has become known as 'Evolutionary Machine Learning'. After setting the context, we shall look in particular at Genetic Programming, one of the core areas of evolutionary machine learning. The talk will introduce its basic ideas, discuss the current status and its challenges.

Deploying Real Systems to Counter Misinformation in Brazil 

Tuesday, September 26th, 11h40

Abstract: The political debate and electoral dispute in the online space during the 2018 Brazilian elections marked the beginning of a major information war in Brazil. This war has become part of our daily lives and one of the most challenging problems in our society. In order to mitigate the problem, we created the project “Elections without Fake” ( and we deployed technological solutions capable of monitoring and exposing the actions of different political campaigns in the online space. Examples of systems include: a monitor for advertisements on Facebook and monitors for public groups, focused on political discussion, on WhatsApp and Telegram. Our systems have proven to be fundamental for fact-checking, for investigative journalism, becoming a partner of the Superior Electoral Court as part of the national front to combat misinformation. This talk summarizes a few lessons learned from the deployment of these systems and points to future directions for combating misinformation.

Ricardo Baeza-Yates, Institute for Experiential AI @ Northeastern University

Responsible AI

Tuesday, September 26th, 14.00

Abstract: In the first part, to set the stage, we cover irresponsible AI: (1) discrimination (e.g., facial recognition, justice); (2) phrenology (e.g., biometric-based predictions); (3) limitations (e.g., human incompetence, minimal adversarial AI) and (4) indiscriminate use of computing resources (e.g., large language models). These examples do have a personal bias but set the context for the second part where we address three challenges: (1) principles & governance, (2) regulation and (3) our cognitive biases. We finish discussing our responsible AI initiatives and the near future.

Autonomous Machine Learning (AutoML): evolving neural network models with quantum-inspired evolutionary algorithm

Wednesday, September 27th, 9.00

Abstract: The area of ​​autonomous machine learning (AutoML) aims to develop decision support systems automatically. The goal is to make machine learning accessible to other scientists who want to apply these techniques to their domains. The area of ​​neuro-evolutionary models or, more recently, Neural Architecture Search (NAS), can be seen as a sub-area of ​​AutoML and is an essential step towards automating the development of neural network architectures. This talk presents an overview of quantum-inspired evolutionary models and their application in the automatic evolution of different models of artificial neural networks, from simple (shallow) Multi-Layer Perceptrons and Recurrent Neural Networks to more complex deep Convolutional Neural Networks. Applications in control, system identification, image classification and time series forecasting will be presented.

Marco Aiello, University of Stuttgart

Data is nothing without control: On building IoT Architectures for a Sustainable Society

Wednesday, September 27th, 14.00

Abstract: The availability of data in digital form is at unprecedented levels and is enabling a new way of building systems that are transforming our society. But data by itself is not enough to provide for useful systems. In this talk, I will overview recent and current research efforts at the Service Computing Department of the Institute of Architecture of Application Systems of the University of Stuttgart. The research efforts revolve around the design of architectures that go beyond the mere collection of data and are able to support humans in smart ways. The applications of these systems are in the field of energy and sustainability, in particular: Smart Buildings, Smart Grids, and Smart Data Centers. 

Frank Hutter, University of Freiburg

Deep Learning 2.0: Towards AI that Builds and Improves AI

Thursday, September 28th, 9.00

Abstract: Throughout the history of AI, there is a clear pattern that manual elements of AI methods are eventually replaced by better-performing automatically-found ones; for example, deep learning (DL) replaced manual feature engineering with learned representations. The logical next step in representation learning is to also (meta-)learn the best architectures for these representations, as well as the best algorithms & hyperparameters for learning them. In this talk, I will discuss various works with this goal in the area of AutoML, highlighting that AutoML can be efficient and arguing for an emphasis on multi-objective AutoML to also account for the various dimensions of trustworthiness (such as algorithmic fairness, robustness, and uncertainty calibration). Finally, taking the idea of meta-learning to the extreme, I will deep-dive into a novel approach that learns an entire classification algorithm for small tabular datasets that achieves a new state of the art at the cost of a single forward pass.

Felipe Trevizan, Australian National University

Planning under Uncertainty: new results and applications

Thursday, September 28th, 11.40

Abstract: The level of autonomy of computational and robotics systems is rapidly increasing, from smart grid agents to autonomous vehicles. These systems are required to reason about their actions in order to achieve their designed complex goals while handling the uncertainty posed by their environment and the results of their own actions.  Although progress has been made to solve some of these problems, for instance, autonomous cars are being tested under real-world conditions, the developed solutions remain domain-dependent, that is, the techniques developed there might not be applicable to other domains such as smart grid agents. It remains a challenge to develop efficient domain-independent methods for solving the different classes of planning under uncertainty problems. In this talk, I will present some of our recent results in developing efficient domain-independent methods for different planning.

Ana Bazzan, UFRGS

Networks, networks, and more networks: applications in humanities, data science, and machine learning

Thursday, September 28th, 14.00

Abstract: It is known that networks or graphs can be used in machine learning and data science to represent and analyze data that has complex relationships. Besides these uses, networks are also relevant to the overall AI agenda in at least two aspects. First, it relates to automated data gathering and language models in the semantic web, since the actual data have to be acquired in some manner in order to form the graphs. Second, it can be used to accelerate learning tasks, as in the case of reinforcement learning. In this talk I present examples of how data is acquired and used in applications in the Humanities (history, storytelling) in order to discover patterns and/or to investigate assumptions. Then, I discuss applications on data science and machine learning, as for instance the use of networks in reinforcement learning, with examples from urban mobility and car to infrastructure communication.