Unisi-IA

News

-12th of April 2024, 10 am seminar by Fabio Zanasi is Professor of Computer Science at University College London, UK, and also holds an appointment at University of Bologna, Italy. His research develops algebraic and logical methods for the compositional analysis of computing systems. He works on the foundations of concurrency, cyber-physical systems, probabilistic programming, and machine learning.

 Abstract

I will present recent and ongoing work on giving a semantic foundation to training algorithms in machine learning using the categorical formalisms of lenses. Lenses provide a much needed unifying perspective on various classes of such algorithms, as well as offering a different style of specifying and proving properties of training protocols. They also enable the study of machine learning for new classes of models such as Boolean circuits and polynomial circuits.

-18th January 2024, 11 am Seminar by Michelangelo Diligenti and Francesco Giannini, from the DIISM (Universitá di Siena), both in presence in the lab 201 (DIISM) and online. 

Abstract: 

This talk delves into the neural-symbolic reasoning paradigm, aiming to bridge the gap between the power of neural networks and the interpretability required for real-world applications. While neural networks excel at capturing complex patterns in data, their opaque nature poses challenges for understanding their decision-making processes. The presentation explores novel approaches that integrate symbolic reasoning into neural architectures, enhancing interpretability by design. By combining the strengths of symbolic logic and neural networks, we aim to create models that not only achieve high performance but also provide insights into their decision-making, opening avenues for applications in critical domains. In particular, we will discuss how neural-symbolic reasoning methods can be instantiated in relational domains, like for large knowledge graphs,  and discuss some implications for building more transparent, accountable, and trustworthy AI systems.

*******************

-20th December 2023, 12 pm, we will have the last Unisi-IA seminar of 2023. 

The speaker will be Federico Becattini, researcher of the SAILAB group (Dipartimento di Ingegneria dell'Informazione e Scienze Matematiche). The seminar will be both in presence in the lab 201 of the DIISM and online.

Title - Analyzing Faces with Neuromorphic Vision

Abstract - Neuromorphic vision is a bio-inspired vision paradigm where, in contrast with traditional vision systems, instead of generating and processing synchronous streams of frames, asynchronous streams of pixel-level events are obtained. Such a paradigm is enabled by event cameras, devices that can capture local illumination changes (events) at an extremely fast rate, with a temporal precision that reaches microseconds. Many fields of research have found effective use of neuromorphic sensors, ranging from slam to drone vision, human understanding and automotive. In this talk, we will explore the possibility of exploiting event cameras to analyze faces. In particular, thanks to the low latency of event cameras, we are interested in capturing facial micro-expressions that normally occur within a few milliseconds and that can be directly tied to micro-movements of specific facial muscles. Nonetheless, working with neuromorphic cameras for facial analysis has its shortcomings. Decades of research on RGB images and videos cannot be fully leveraged in the event domain and labeling data is costly and time-consuming. We will present an overview of the opportunities and challenges that can be found in this domain and explore the few existing solutions available in the state of the art.

-17th November 2023, 11 a.m. Seminar by Prof. Pietro Rubegni and Dr. Alessandra Cartocci (Department of Medicine, Unisi). Title: "Position statement of the EADV Artificial Intelligence (AI) Task Force on AI-assisted smartphone apps and web-based services for skin disease" . This time the seminar will only take place online at the link disseminated through the mailing list. 

-17th October 2023, 11 a.m., Seminar Prof. Antonio Rizzo group both in Lab 201 of the San Niccolò building and online.

Title: Reinforcement Learning for Process Control

Speakers: Dr. Leonardo Guiducci and Prof. Antonio Rizzo (DISPOC, Unisi)

Abstract: The origin of the Reinforcement Learning (RL) paradigm in machine learning can be traced back to the early works in psychology and optimal control theory. The two threads started to be more closely related with the development of Deep Q-Learning, leading to the modern reinforcement learning algorithms we see today. We present a few examples of our explorations of RL for process control  inspired by cognitive processes and conclude with our combination of value-based and policy-based methods in the Actor-Critic approach to Reinforcement Learning. This approach incorporates MILP Optimal Control, as demonstrated in a case study focused on Energy Management within MicroGrids.

-18 April 2023, 11 a.m., Seminar Prof. Marco Corneli (https://math.unice.fr/~mcorneli/) 

Title: Artificial Intelligence for historical and archaeological data analysis

Abstract: In this talk Prof. Corneli will present some of the ongoing projects my team is involved in at the laboratory CEPAM at Université Côte d'Azur, Nice France. Historical and archaeological data comes from a variety of different sources, mostly involving text (from painted images, inscriptions, ancient books or archives) and images (painted, microscopical, tomographic, 2D or 3D). Despite the huge heterogeneity of the sources, the historical/archaeological data comes with a few common features : it is scarce, incomplete and dramatically heterogeneous. These features raise a number of key questions in AI and machine learning approaches, related with few shot learning, missing data imputation, iner-active learning and the development of ad-hoc neuro-symbolic architectures capable of taking into account the domain experts' knowledge. Some relevant case studies, first results and future challenges will be discussed.  

-27 Marzo 2023: ore 15-17

Workshop su Etica e Intelligenza Artificiale

Sarà possibile seguire l'iniziativa in presenza, presso il Rettorato, e online, all'indirizzo che vi sarà inviato qualche giorno prima.

Per partecipare vi preghiamo di iscrivervi a questo indirizzo 

https://forms.gle/PAiYbTRV45sinp1Q6 

entro il 25 marzo 2023.

Tutti i Dettagli nella Locandina seguente.