Remote Sensing of Biophysical Parameters in Inland Waters
Milad Niroumand Jadidi
Wednesday 16 Feb 2022 - on-line
abstract
The biophysical attributes of inland waters (e.g., bathymetry and water quality) are closely linked to various aquatic ecosystem services such as drinking water, aquaculture, and recreation. Satellite remote sensing provides an efficient means of characterizing these parameters across large spatial and temporal extents. This talk will provide an overview of optical remote sensing over water bodies, focusing on physics-based aspects and challenges, e.g., confounding effects from the atmosphere. Then, advanced physical and machine learning-based methods will be addressed for automatically mapping a set of key in-water and benthic parameters. The retrieved parameters include hydro-morphological attributes of fluvial systems such as bathymetry as well as multitemporal dynamics of lake water quality parameters (e.g., chlorophyll-a and total suspended matter).
bio
Milad joined Remote Sensing for Digital Earth (RSDE) unit of FBK as a researcher at the end of 2017. He received a joint Ph.D. degree in Civil and Environmental Engineering from University of Trento, Italy, and Freie Universität Berlin, Germany, in 2017. His research is focused on retrieving biophysical parameters in inland and coastal waters from optical imagery.
The Problem of Bias in Hate Speech Detection
Alan Ramponi
Wednesday 2 Feb 2022 - on-line
abstract
Hate speech in online social communities is a serious and pervasive concern, which requires fair and robust automated approaches to be tackled at scale. However, despite the great progress in natural language processing for detecting hate speech, current models have shown to be brittle when applied to real-world data, exhibiting limited out-of-distribution robustness and perpetuating and amplifying harmful social biases. After a brief overview on online hate speech and the multifaceted meaning of "bias" in natural language processing, we will dig into the topic, critically analyzing lexical biases in hate speech detection via a cross-platform study and analyzing their impact on out-of-distribution robustness and fairness. We conclude by discussing future avenues for research, arguing that fairness and robustness are strongly intertwined aspects that should be studied jointly in future work.
bio
Alan Ramponi is a postdoctoral researcher in natural language processing (NLP) at Fondazione Bruno Kessler in the Digital Humanities (DH) research group. His research interests include transfer learning, language variation, bias and fairness in NLP. More broadly, he is interested in making NLP robust and more fairly applicable to social sciences problems. Before joining FBK, he was a visiting researcher in the NLPnorth group at the IT University of Copenhagen, Denmark. He earned his Ph.D. and his M.Sc. in computer science from the University of Trento.
Digital proximity tracing on empirical networks for pandemic control
Giulia Gencetti and Gabriele Santin
Wednesday 19 Jan 2022 - on-line
abstract
Digital contact tracing is a relevant tool to control infectious disease outbreaks, including the COVID-19 epidemic. Early work evaluating digital contact tracing omitted important features and heterogeneities of real-world contact patterns influencing contagion dynamics. We fill this gap with a modeling framework informed by empirical high-resolution contact data to analyze the impact of digital contact tracing in the COVID-19 pandemic. We investigate how well contact tracing apps, coupled with the quarantine of identified contacts, can mitigate the spread in real environments. We find that restrictive policies are more effective in containing the epidemic but come at the cost of unnecessary large-scale quarantines. Policy evaluation through their efficiency and cost results in optimized solutions which only consider contacts longer than 15–20 minutes and closer than 2–3 meters to be at risk. Our results show that isolation and tracing can help control re-emerging outbreaks when some conditions are met: (i) a reduction of the reproductive number through masks and physical distance; (ii) a low- delay isolation of infected individuals; (iii) a high compliance. Finally, we observe the ineffi- cacy of a less privacy-preserving tracing involving second order contacts. Our results may inform digital contact tracing efforts currently being implemented across several countries worldwide.
bio
Giulia Cencetti has been a researcher in the MobS group (Mobile and Social Computing Lab) at FBK for more than a year now, where she has been dealing with network science, models of social interactions, urban planning, and other applications to the social sciences of complex systems. In her work she integrates her previous studies (degree in physics and a doctorate on complex systems) with computer and data science techniques. Acquiring these skills was possible thanks to various research experiences, such as at Queen Mary University of London and Central European University, and above all thanks to the multidisciplinary and intercultural wealth of the MobS research group led by Bruno Lepri.
Gabriele Santin has been a researcher at FBK since 2019. Before joining the MoBS group of Bruno Lepri he obtained a doctorate in computational mathematics from the University of Padua, and was subsequently postdoc at the University of Stuttgart, dealing with machine learning also within the Cluster of Excellence on Simulation Science. In the MoBS group he works mainly on machine learning and data science problems, with a growing interest in network science and computational social science applications.
Cross-Modal Knowledge Transfer via Inter-Modal Translation and Alignment for Affect Recognition
Vandana Rajan
Wednesday 15 Dec 2021 - on-line
abstract
Multi-modal affect recognition models leverage complementary information in different modalities to outperform their uni-modal counterparts. However, due to the unavailability of modality-specific sensors or data, multi-modal models may not be always employable. For this reason, we aim to improve the performance of uni-modal affect recognition models by transferring knowledge from a better-performing (or stronger)modality to a weaker modality during training. Our proposed multi-modal training framework for cross-modal knowledge transfer relies on two main steps. First, an encoder-classifier model creates task-specific representations for the stronger modality. Then, cross-modal translation generates multi-modal intermediate representations, which are also aligned in the latent space with the stronger modality representations. To exploit the contextual information in temporal sequential affect data, we use Bi-GRU and transformer encoder. We validate our approach on two multi-modal affect datasets, namely CMU-MOSI for binary sentiment classification and RECOLA for dimensional emotion regression. The results show that the proposed approach consistently improves the uni-modal test-time performance of the weaker modalities.
bio
Vandana Rajan received the Master of Technology degree in Digital Signal Processing from the Indian Institute of Space and Technology in 2016 and worked for 2 years as an embedded software engineer. She is currently pursuing her Ph.D degree at the Queen Mary University of London, supervised by Prof. Andrea Cavallaro and Dr. Alessio Brutti. Her research interests mainly include multi-modal signal processing, computational paralinguistics, and speech emotion recognition.
Human-in-the-Loop for Data Collection: a Multi-Target Counter Narrative Dataset to Fight Online HateSpeech
Helena Bonaldi
Wednesday 1 Dec 2021 - on-line
abstract
Undermining the impact of hateful content with informed and non-aggressive responses, called counter narratives,has emerged as a possible solution for having healthier online communities.Thus, some NLP studies have started addressing the task of counter narrative generation. Although such studies have made an effort to build hate speech /counter narrative (HS/CN) datasets for neural generation, they fall short inreaching either high-quality and/or high-quantity. In this seminar, we presenta novel human-in-the-loop data collection methodology in which a generative language model is refined iteratively by using its own data from the previous loops to generate new training samples that experts review and/or post-edit.Our experiments comprised several loops including diverse dynamic variations.Results show that the methodology is scalable and facilitates diverse, novel,and cost-effective data collection.
bio
Helena Bonaldi is a first year Ph.D.student in the LanD group, under the supervision of Marco Guerini. Her research focuses on dialogical models for counter narrative generation to fight hate speech. Before starting her Ph.D., she was a Data Science student at theUniversity of Trento, and an internship student at Fondazione Bruno Kessler(FBK).
Integrating Planning, Acting, and Learning in Unknown Environments
Leonardo Lamanna
Wednesday 17 Nov 2021 - on-line
abstract
An Artificial Intelligence agent should be able to accomplish different tasks in unknown environments. One way to achieve this is by interleaving learning, planning, and acting. In many real scenarios, an agent has to operate in unknown, dynamic and complex environments that it can perceive only through a set of sensors. However, to apply symbolic planning, it requires a symbolic representation of its current state. Therefore an agent has to learn such a symbolic representation by mapping continuous perception variables into high-level discrete ones. Next, given a goal, it can compute a sequence of symbolic actions to achieve the goal. Finally, each symbolic action needs to be compiled down into a sequence of low-level operations executable by the agent actuators.
bio
Leonardo Lamanna graduated in bachelor computer engineering on 13th September 2017 with the final grade of 90/110 at the University of Brescia. His bachelor's Thesis was about applying an optimization metaheuristic to solve a NP-hard problem. Then he graduated in master computer engineering on 11th September 2019 with the final grade of 110/110 cum laude at the University of Brescia. His master’s Thesis was about integrating Mathematical Programming and Artificial Intelligence techniques to solve integer linear optimization problems.
Currently, he is a Computer Science Ph.D. student of the University of Brescia and the Data and Knowledge Management (DKM) unit of the Fondazione Bruno Kessler (FBK) research center in Trento. His Ph.D. topic is the integration of learning, planning and acting in unknown, complex, and dynamic environments.
Gamified Education solutions in Software Programming and Modelling: A Tool implementation Perspective
Antonio Bucchiarone
Wednesday 3 Nov 2021 - on-line
abstract
There is a wide interest in adopting gamification solutions for supporting engagement in software modeling and production with learning purposes, e.g., students to actively participate in courses, employees to deal with tedious tasks, or programmers to learn specific languages. Several frameworks to build-up gameful applications have been proposed. Their common goal is raising the level of abstraction of gamification mechanisms and proposing a well-defined set of languages for designing a game, its main components, and the behavioral details.
All the research introduced until now reinforce the idea that there is a strong interest in the use of gamification in modeling and programming. Some early promising results exist, but in general existing attempts are mainly the work of modelers trying to manually create ad hoc gamification environments for their specific experimental scenarios. A real attempt to integrate full gamification in modeling and programming tools is still missing.
Our vision proposes to integrate gamification mechanisms with modeling/programming tools. In this talk I present the required components and the necessary interconnections to adequately assemble altogether and get the resulting gamified modeling/programming tool. Moreover, I present two concrete examples of our idea: I first present an example of a gamified software, Polyglot, conceived to be applied in the field of software programming. After this I introduce PapyGame, a gamified software which finds its application in the field of software modeling. Finally, I illustrate a list of research challenges that must be addressed to realize motivational digital systems able to support personalized learning plansand feedback in context as education and training.
bio
Antonio Bucchiarone is currently a Senior Researcher at the Motivational Digital Systems (MoDiS) research unit. His research activity is focused principally on many aspects of the Software Engineering for Adaptive Socio-Technical Systems. In the last 12 years, he has investigated advanced methodologies and techniques supporting the definition, development, and management of distributed systems that operate in dynamic environments, where being adaptable is a key intrinsic characteristic.
Assessing infodemics in complex socio-technical systems
Riccardo Gallotti
Wednesday 20 Oct 2021 - on-line
abstract
An infodemic is the overabundance of (dis)information running across our contact networks because we fail at distinguishing reliable information from unreliable "fake news". Thanks to the Covid19 Infodemic Observatory, we are able to capture and process an unprecedented amount of data describing the widespread infodemic associated with the COVID19 pandemic.
In this seminar, I describe how we harness these data to understand the patterns of infodemiological behaviour. I will offer an overview ranging from our earliest results, describing the social mechanics of the infodemics-epidemics interaction at a national level, to our last efforts, focusing on the behaviour of individual by examining not only the communications' informative (or dis-informative) content, but also their intent, measured as the level of hate speech classified with Google's BERT model.
bio
Riccardo is an interdisciplinary physicist working on the data-informed statistical modelling of individual and collective behaviors in the CoMuNe Lab of the Bruno Kessler Foundation. His research focuses on human mobility, decision making, transportation, infodemics, and data science. Before joining the Foundation, Riccardo was affiliated to the Institute for Interdisciplinary Physics and Complex Systems in Mallorca (Spain), the Center for Complex Systems and Brain Sciences in San Martin (Argentina) and the Institute for Theoretical Physics of the Atomic Energy Committee in Saclay (France).
User-centred design in the wild: designing technology for mountain sports
Eleonora Mencarini
Wednesday 6 Oct 2021 - on-line
abstract
Mountain sports are based on the challenge of a natural element like the verticality of rock walls, the deepness of caves, or the unpredictability of snow. Their performance requires physical preparation, psychological and emotional firmness, coordination with partners, and environmental knowledge. In particular, regarding the latter aspect, outdoor sportspeople can offer a new perspective on the territory and its environmental protection. In this seminar, I will present my research on climbing and speleology, highlighting what user-centered design (and the Digis center at large) can do for sportspeople communities and what they can do for us.
bio
Eleonora Mencarini is a researcher at the i3 research unit of FBK. She holds a PhD (cum laude) in Computer Science from the University of Trento (Italy) and an MA (cum laude) in Communication Science from the University of Siena (Italy). Her main research interests are ethnography, co-design, embodied interaction, and interaction design in natural environments. The main application fields where she has conducted her research are outdoor sports, wearable devices, citizen science, playful education.
Continual Learning in Image Classification and Semantic Segmentation
Enrico Fini
Wednesday 22 Sep 2021 - on-line
abstract
Continual Learning (CL) aims to develop agents emulating the human ability to sequentially learn new tasks while being able to retain knowledge obtained from past experiences. Due to the relevance of its applications, in the last few years, the computer vision research community has put considerable effort into developing CL methods.
After a brief overview of the challenges and the opportunities of CL, we will go through the main families of techniques that are used to mitigate catastrophic forgetting (i.e. the tendency of deep neural networks to forget previously learned information upon learning new information). In addition, we will discuss CL in some interesting applications, as for instance when no information can be transferred between tasks (memory-constrained), and for pixel-wise prediction (semantic segmentation).
bio
Enrico is a Ph.D. student at the University of Trento in the MHUG group, under the supervision of Elisa Ricci. He is currently a visiting researcher in the THOTH team of INRIA (Grenoble, France). His research mainly focuses on continual, self- and semi-supervised learning. Before joining the University of Trento, he was an intern at the European Space Agency (ESA) and Fondazione Bruno Kessler (FBK). He received his master's degree from Politecnico di Milano and his bachelor's degree from the University of Parma.
Tiny architectures for tiny architectures
Francesco Paissan
Wednesday 2 Mar 2022 - on-line
abstract
In the Internet of Things (IoT) era, where we see many interconnected and heterogeneous mobile and fixed smart devices, distributing intelligence from the cloud to the edge has become critical for the sustainability of the infrastructures and comes with additional benefits (e.g. power efficiency, low-latency inference, privacy-by-design) with respect to centralized cloud computing. TinyML is an emerging field that has gained much traction in the last few years. In particular, this paradigm brings novel challenges in the deep learning domain, such as low memory availability and limited energy budget. In this seminar, I will summarize the work done to tackle the challenges of tinyML with novel neural architectures and training paradigms. Our main focus is on multimedia analytics at the edge and privacy-by-design systems. Our efforts landed in state-of-the-art audio and video analytics performance, namely in sound event detection and multi-object detection and tracking with a fraction of the computational complexity of the previous state-of-the-art approaches. Moreover, we developed a novel approach to generate deep-fakes for privacy-by-design applications directly at the edge. Part of the work presented regards the intra-DigiS collaboration between E3DA and SpeechTek Lab and the EU project MARVEL.
bio
Francesco Paissan joined the Energy Efficient Embedded Digital Architectures (E3DA) unit in Fondazione Bruno Kessler (FBK) in 2018 as a junior researcher. His research interests are shared among diverse topics, ranging from developing and modeling scalable neural architectures for multimedia analytics to bio-signals analysis with deep learning architectures. In 2021, Francesco joined the LEGEND experiment for the design of novel physics-inspired ML algorithms (e.g. learning-based triggering logics for cosmogenic rejection in the experiment's veto). Moreover, Francesco is actively contributing to the ENVISION project, funded by the European Space Agency and to the Speechbrain toolkit. Last but not least, Francesco is currently in the final semester of his BSc degree in Physics at the University of Trento.
Emergence of Topological Shortcuts in Machine Learning
Giulia Menichetti
Wednesday 16 Mar 2022 - on-line
abstract
In recent years deep learning models have become popular tools to perform link prediction tasks in presence of network structures rich in metadata. Common deep learning frameworks formulate link prediction as a binary classification task, connecting node entities according to their features. The successful training of a binary classifier requires node pairs that are known to bind to each other, as well as negative samples, i.e., pairs that do not interact. Such positive and negative records are usually determined by thresholding continuous variables characterizing the strength of the interaction between two nodes. However, we often observe that link strength is not randomly distributed across the records, but there is correlation between number of annotations and average interaction strength per node. As the annotations commonly follow fat-tailed distributions, the observed correlation drives the hub nodes to have disproportionately more positive links on average, whereas nodes with fewer annotations have more negative examples. Uniformly sampled training datasets affected by this annotation imbalance prompt ML models to learn and predict that some nodes are connected disproportionally more often than others. In other words, ML models learn the connection patterns from the degree of the nodes, neglecting relevant node features. This annotation imbalance offers good performance for the unknown annotations associated with the missing links in the network used for training, a phenomenon we term emergence of topological shortcuts. A key consequence and a signal of such topological shortcuts is the degradation of the performance of an ML model when asked to perform link prediction between novel (i.e., never-before-seen) nodes.
In this talk, I will cover some of the strategies we have developed to control for over-fitting and annotation imbalance, and maximize generalization to unseen nodes.
bio
Dr. Giulia Menichetti is a Senior Research Scientist at the Network Science Institute (Northeastern University), and an Associate Researcher at Brigham and Women’s Hospital (Harvard Medical School). She is a statistical/computational physicist by training, and during her Ph.D. she specialized in Network Science. She currently leads the Foodome project, which aims to track the full chemical complexity of the food we consume and develop quantitative tools to unveil, at the mechanistic level, the impact of these chemicals on our health.
Orchestration at the Edge: a Key Enabler for 5G and 6G Networks
Rasoul Behravesh
Wednesday 30 Mar 2022 - on-line
abstract
The ongoing roll-out of 5G networks paves the way for many fascinating applications such as virtual reality (VR), augmented reality (AR), and autonomous driving. Moreover, 5G enables billions of devices to transfer an unprecedented amount of data at the same time. This transformation calls for novel technologies like multi-access edge computing (MEC) to satisfy the stringent latency and bitrate requirements of these applications. The main challenge pertaining to MEC is that the edge MEC nodes are usually characterized by scarce computational resources compared to the core or cloud, arising the challenge of efficiently utilizing the edge resources while ensuring that the service requirements are satisfied. When considered with the users’ mobility, this poses another challenge, which lies in minimization of the service interruption for the users whose service requests are represented as service function chains (SFCs) composed of virtualized network functions (VNFs) instantiated on the MEC nodes or on the cloud. In this talk, I will present some of the research studies we have conducted to develop SFC placement and VNF lifecycle management strategies at the edge of mobile networks.
bio
Rasoul Behravesh is a researcher in the Smart Networks and Services (SENSE) unit at Fondazione Bruno Kessler (FBK) in Trento, Italy. Currently, his main research focus is on zero-touch network and service management and orchestration in 5G networks. He holds a Ph.D. in Telecommunications from the University of Bologna, Italy. Prior to that, he received his M.Sc. in Computer Networking from QIAU and his B.Sc. degree in Information Technology from the Payam-e-Noor University of Mahabad. His main research interests include 5G networks, multi-access edge computing, network function virtualization, and network slicing.
AI-Based Solutions for Adaptive Gamified Motivational Systems
Mauro Scanagatta
Wednesday 27 Apr 2022 - on-line
abstract
GMS (Gamified Motivational Systems) is a broad term to indicate the introduction of "game-like" elements with the objective of promoting user engagement, motivation and positive behavior change. Typical examples of those applications are Serious Games, Playful education, Gamification. We will first present the tangible application benefits and the key research challenges related to GMS. We will then show three real-life applications implemented by Digis; for each one we'll detail the application context, the problems encountered, and the specific AI-based solution put into practice.
bio
Mauro is a Researcher at Fondazione Bruno Kessler (FBK), Italy. His background is rooted in statistical models, and his research focuses on the application of Artificial Intelligence to Motivational games, ranging from personalization of game elements to the calibration of game dynamics. His research interest is on how technology can provide more meaningful and profound changes on society through serious games' medium.
Quantifying the robustness of empirical networked complex systems
Oriol Artime Vila
Wednesday 11 May 2022 - on-line
abstract
Power grids, trophic relations, online micro-blogging services or scientific collaborations: complex networks are ubiquitous in nature. In the research agenda of network science, network robustness has been a topic of paramount importance due to its conceptual, modelling challenges and its wide range of applications. In this talk I will review some of our latest research on this direction. I will show how to quantify robustness when networked systems are perturbed in two different scenarios. The first one considers metadata defined on the nodes, so metadata-informed attacks are performed in order to understand how to efficiently dismantle (or protect) a network. Thus, we shed light on the importance of non-toological information in network interventions. The second one focuses on the under-researched but empirically relevant phenomenon of non-local cascade spreading. In this, non-local cascading failures evolve on a network in a way that consecutive node failures need not to strictly follow a proximity-based evolution. We propose a mathematically solvable model of such a cascades and evaluate their accuracy in artificial and empirical topologies, finding an excellent agreement.
bio
Oriol Artime Vila is a postdoctoral researcher at CHuB Lab. He holds a PhD in Physics from the Institute for Cross-Disciplinary Physics and Complex Systems (UIB-CSIC, Spain). His research interests lie at the interface between complexity science and network theory, and in the last years he has devoted a great deal of attention to problems related to the robustness of networked systems.
Automatic View-Specific Assessment of L2 Spoken English
Stefano Bannò
Wednesday 8 Jun 2022 - on-line
abstract
The growing demand for learning English as a second language has increased interest in automatic approaches for assessing and improving spoken language proficiency. A significant challenge in this field is to provide interpretable scores and informative feedback to learners through individual viewpoints of learners’ proficiency, as opposed to holistic scores. Thus far, holistic scoring remains commonly applied in large-scale commercial tests. As a result, an issue with more detailed evaluation is that human graders are generally trained to provide holistic scores.
This talk investigates whether view-specific systems can be trained when only holistic scores are available. To enable this process, view-specific networks are defined where both their inputs and structure are adapted to focus on specific facets of proficiency. It is shown that it is possible to train such systems on holistic scores, such that they provide view-specific scores at evaluation time. View-specific networks are designed in this way for pronunciation, rhythm, text, use of parts of speech and grammatical accuracy. The relationships between the predictions of each system are investigated on the spoken part of the Linguaskill proficiency test. It is shown that the view-specific predictions are complementary in nature and capture different information about proficiency.
bio
Stefano Bannò is a 3rd year PhD student in Cognitive Science at University of Trento in a joint programme with Fondazione Bruno Kessler (FBK) and is working on a project on automatic assessment of spoken language proficiency. As a part of his PhD, he has worked at the ALTA (Automated Language Teaching and Assessment) Institute of the Department of Engineering at Cambridge University as a visiting student. Before starting the doctoral course, he obtained a master’s degree in Philology and a bachelor’s degree in Classics at University of Trento. During his master’s studies, he spent a research period at the Lautarchiv of Humboldt University in Berlin.
Besides his academic career, he has worked as a musician and a secondary school teacher. His research interests span from machine learning and natural language processing to phonetics and sociolinguistics.
Neighbourhood matching creates realistic surrogate temporal networks
Antonio Longa
Monday 20 Jun 2022 - on-line
abstract
Temporal networks are essential for modeling and understanding systems whose behavior varies in time, from social interactions to biological systems.
Often, however, real-world data are prohibitively expensive to collect or unshareable due to privacy concerns.
A promising solution is 'surrogate networks', synthetic graphs with the properties of real-world networks.
Until now, the generation of realistic surrogate temporal networks has remained an open problem, due to the difficulty of capturing both the temporal and topological properties of the input network, as well as their correlations, in a scalable model.
Here, we propose a novel and simple method for generating surrogate temporal networks.
By decomposing graphs into temporal neighborhoods surrounding each node, we can generate new networks using neighborhoods as building blocks.
Our model vastly outperforms current methods across multiple examples of temporal networks in terms of both topological and dynamical similarity.
We further show that beyond generating realistic interaction patterns, our method is able to capture intrinsic temporal periodicity of temporal networks, all with an execution time lower than competing methods by multiple orders of magnitude.
bio
Antonio is a PhD student at the Fondazione Bruno Kessler and the University of Trento, under the supervision of Bruno Lepri and Andrea Passerini. He is currently a visiting researcher at Cambridge University (Cambridge, UK). His research mainly focuses on Temporal Graph Mining and Geometric Deep Learning. Before joining the Fondazione Bruno Kessler, he studied as an exchange student at Aalto University (Finland) and he did his master thesis at the University of Exeter (UK). He received his master's degree from the University of Trento and his bachelor's degree from Milano-Bicocca University.
Person kicking ball: When Vision meets Language
Adrian Muscat
Wednesday 21 Sep 2022 - Fondazione Bruno Kessler
abstract
The detection of relationships between objects in images is a sub-task in joint computer vision and language research and finds application in for example Visual Question Answering and Image Caption Generation as applied to robotics. Following an introduction to the topic, various models, including deep neural networks and classical machine learning methods, that explicitly or implicitly detect or generate relations are described followed by an analysis of where the models fail. The talk is concluded with an in-depth discussion on potential solutions.
bio
Adrian Muscat holds the position of professor at the Department of Communications and Computer Engineering, University of Malta. He received the Ph.D. in Electronics Engineering from Queen Mary University of London in 2002, the M.Sc. Degree in RF and Microwave Engineering from the University of Bradford in 1996, and the B.Sc. degree in Electrical Engineering from the University of Malta, 1993. During the early stages of his career he carried out research mainly in computer aided design, during which time he studied and adapted evolutionary search techniques for 2D shape optimisation and developed a modified line grammar to generate microstrip antenna shape candidates, which are typical of what human designers develop during the early phase design. His current research interests are in computer vision and image understanding, with special emphasis on detecting spatial relations in images and video and their application in robotics, autonomous vehicles and assisted living.
Promoting Artificial Intelligence Literacy in Young People and in Educators
Gianluca Schiavo
Monday 10 Oct 2022 - on-line
abstract
Artificial intelligence is gaining an increasing presence in our lives, and it will only grow. In this talk, I will address the increasing role that artificial intelligence is playing in the lives of young people, especially in the educational domain. The exposure of young generations to AI via smartphones, streaming platforms, videogames, and social networks suggests a world in which algorithms play a ubiquitous role. However, society and the educational system are doing little to equip young people and educators with the skills to navigate this world with agency and literacy. I present our contribution to fostering and enriching AI literacy for everyone, especially young generations and K-12 teachers, by adopting multidisciplinary approaches and targeting diverse populations. The overall goal is to advance knowledge and technologies that support how diverse people learn about AI, as well as explore how AI applications can help to better support human learning.
bio
Gianluca Schiavo is a researcher at the i3 research unit at FBK. He holds a PhD in Cognitive Science from the University of Trento (Italy) and a MSc in Psychology from the University of Padova (Italy). His research centers on Human-Computer Interaction (HCI), focusing on the design, development, and evaluation of collaborative, social, and accessible technologies.
His current research is at the boundary of HCI and Artificial Intelligence to explore issues surrounding AI literacy and human-AI interaction, especially in the educational domain.
Proximity Detection: Energy Efficient and Accurate (and not just for COVID Contact Tracing)
Amy Lynn Murphy
Wednesday 19 Oct 2022 - Fondazione Bruno Kessler and on-line
abstract
We all became familiar with proximity detection during the pandemic. Some of us even installed "contact tracing" apps, with proximity detection at their core. While contact tracing on smartphones was ubiquitous, it had other drawbacks. First and foremost, it wasn't very accurate - a direct consequence of using Bluetooth to "measure" the distance.
In this talk, I will introduce Janus, our contact detection protocol that is both accurate and energy efficient. I will show how reaching inside the communication protocol stack allows us to fully exploit a combination of Bluetooth and ultra-wideband radios for a contact detection protocol that can be used for collecting contact tracing logs as well as for proximity warning systems. I'll also talk about several real-world experiments with Janus in 2020 including two here at FBK (one in the cafeteria and one with 90 people all around the Povo campus) and another with kids attending Trentino summer camps. I'll also provide evidence for the accuracy and energy efficiency of Janus.
bio
Amy L. Murphy is a researcher in the Energy Efficient Embedded Digital Architectures (E3DA) unit at the Fondazione Bruno Kessler in Trento, Italy. Her research focuses on applied research for smart cities from the software engineering, distributed computing, and low-power wireless networks, with a recent emphasis on the Internet of Things. The theme that drives her work is to enable reliable applications for dynamic environments with particular attention to the wireless communication protocols necessary to support complex interactions among distributed devices.
Knowledge rules in support of point cloud semantic segmentation
Alessandro Daniele and Eleonora Grilli
Thursday 3 Nov 2022 - on-line
abstract
Digital Industry & Digital Society Seminar.
Deep Learning approaches have sparked much interest in the AI community during the last decade, becoming state-of-the-art in domains like pattern recognition, computer vision, and data analysis. In the geospatial and remote sensing field, the availability/unavailability of annotated training data is often a big obstacle, and most of the time, we have to deal with small samples and unbalanced classes. To overcome these problems, we introduce KENN (Knowledge Enhanced Neural Networks) within the 3D semantic segmentation pipeline. KENN is a library that allows neural network models to be modified by providing logical knowledge in the form of a set of logic constraints.
To give a practical example, in a point cloud semantic segmentation challenge, the rule which states that "poles"cannot be over a "building" is employed by KENN to avoid misclassification problems with antennas.
The work presented is the result of a collaboration between the 3D Optical Metrology unit and the Data Knowledge and Management (DKM) Unit.
bio
Alessandro Daniele received his master's degree in Computer Science at Università degli Studi di Padova and his Ph.D. at Università degli Studi di Firenze. During his Ph.D. he worked in the Data Knowledge and Management (DKM) group at FBK, where he is currently a researcher in Neural-Symbolic Integration. His research topic is about the integration of logical knowledge with neural networks models, with a particular focus on relational domains.
Eleonora Grilli is a full-time postdoc researcher at the 3D Optical Metrology research group (3DOM) at the Bruno Kessler Foundation (FBK) in Trento. After her master's degree in Civil Engineering at the University of Bologna, she started and completed her Ph.D. at 3DOM under the supervision of Dr. Fabio Remondino, mainly focused on machine learning semantic segmentation applied to architectural and cultural heritage 3D datasets. Her current role involves the design of geometry and machine learning-based solutions in a wide range of fields, such as urban digitisation, digital agriculture and forestry, corridor mapping, and restoration projects.
Un approccio etico alla società digitale
Franco Manti
Monday 7 Nov 2022 - Fondazione Bruno Kessler e on-line
abstract
Le ICT, attraverso le quali si originano nuovi ambienti, hanno portato al superamento della tradizionale contrapposizione fra realtà, possibilità, potenzialità poiché in essi ciò che è possibile si è già realizzato e ciò che sembra essere solo potenziale ha in sé il potere di farsi reale. Perciò, realtà virtuale non è più un ossimoro, ma definisce un diverso livello ontologico rispetto a quello del mondo fisico in cui viviamo nella nostra quotidianità. Questo comporta una forte ricaduta sulle nostre relazioni, ne genera nuove, influisce sulla qualità delle nostre vite e sulla nostra stessa autonomia nel dare (a noi e agli altri) ragione di scelte e comportamenti. Partendo da queste considerazioni, il seminario affronta il problema di definire un approccio etico per la nuova società digitale che si sta imponendo.
Le questioni morali e le prospettive etiche poste dalle relazioni con e nello spazio virtuale ci pongono due quesiti: le realtà virtuali, aumentate, ecc. sono di per sé virtuose? L’approccio con ambienti nei quali reale, potenziale, possibili sono unificati richiede, ancora, la necessità di una distinzione fra i diversi livelli ontologici?
La ridefinizione del rapporto ambiente - comunicazione comporta implicazioni rilevanti sul piano dell’etica pubblica e dell’esercizio della cittadinanza. Alcuni si preoccupano dell'influenza delle piattaforme online sulla democrazia, altri sottolineano il loro potenziale come nuovi forum pubblici accessibili. La domanda cui rispondere è: in che modo le tecnologie dell'informazione possono sostenere o migliorare la cittadinanza, la partecipazione politica e i comportamenti virtuosi senza cadere nell’utopia (pericolosa) della nuova agorà telematica?
bio
Laureato in Filosofia, insegna Etica Sociale, Etica della Comunicazione, Comunicazione Etico- sociale d’Impresa, Filosofia Morale presso la Scuola di Scienze Sociali - Dipartimento di Scienze della Formazione e presso la Scuola di Scienze Mediche e Farmaceutiche - Dipartimento di Scienze della Salute dell’Università di Genova. È Direttore di EtApp - Laboratorio di ricerca per le etiche applicate dell’Università di Genova e dell’ International Research Office for Bioethics Education of the European Centre for Bioethics and Quality of Life – International Chair in Bioethics; è membro dell’Accademia per la finanza sostenibile, di Athena - Network universitario della Fondazione Pubblicità Progresso. Dirige il Master universitario di II livello in Pratiche di filosofia e il Master universitario di I livello Esperto in terapie, attività ed educazione assistite con animali. È membro dell’Ufficio di Gabinetto del Rettore, del Comitato di coordinamento Università di Genova – CCIAA delle Riviere Liguri, del Direttivo del programma europeo Flag – GAC Mare delle alpi.
Ha pubblicato numerosi scritti e monografie su temi di etica, filosofia politica, bioetica.
Globally scaled detection of burned areas with remote sensing
Massimo Zanetti
Wednesday 30 Nov 2022 - Fondazione Bruno Kessler and on-line
abstract
Global scale assessment of the fraction of vegetated land affected by fires, i.e., burned area (BA), is essential for climate related studies, and open-access satellite-borne multi-spectral imagery represents an invaluable source of information for the mapping purpose.
Characterization of BAs through spectral analysis is complex and several factors such as vegetation type, localeco-climate systems, seasonal cycles and temporal attenuation need to be taken into account, leading to the practical impossibility of representing the BA spectral signature as an invariant property at global level and forcing classification algorithms to be extremely dependent on local models that are difficult to train and on a variety of ancillary data that are usually not available at large scales.
In a recent study, these aspects are investigated by focusing the analysis to multi-spectral derived features known as normalized difference indexs (NDIs), a family of band-algebra transofrmations producing variables with known established correlations tophisically-related post-fire effects on vegetation, and a novel classification framework is presented which is aimed at reducing the locality of the BA multi-spectral signature representation while enabling for an improved globally extended general characterization. The effectiveness of the proposed approachis demonstated in a global scale BA detection simulation validated against recently published BA reference global data.
bio
Massimo Zanetti is a researcher in the Remote Sensing for Digital Earth (RSDE) unit at the Fondazione Bruno Kessler (FBK) in Trento, Italy.
His background is on mathematical methods for image analysis and numerical optimization. His reasearch is focused on remote sensing problems such as multi-spectral image analysis and change detection models. The rationale behind his work is to develop models and systems to automatically extract valuable information from last generation EarthObservation (EO) satellite-borne imagery data, with focus on climate change related problems.
Public Administration and Data (Re)use: Dataspaces, Collective Rationality, and their Limits
Riccardo Nanni
Tuesday 13 Dec 2022 - Fondazione Bruno Kessler and on-line
abstract
In this seminar we assess public administrations’ obstacles and best practices in data exposure and reuse. We employee three research methods: “problematisation as a method”, expert interviews, and focus groups. This research contributes to the literature on rationality in decision-making, observing the extent to which data-driven policymaking expands the boundaries of bounded rationality through collective rationality. Concretely, we find that smaller public administrations often lack competences to keep track of existing data within the organisation itself and to make use of it in decision-making. This entails that participation in dataspaces and the use of data in decision-making is more applicable to bigger-size organisations. Therefore, while pooling data in dataspaces expands the boundaries of bounded rationality through collective rationality, limits remain due to the underrepresentation of smaller institutional actors and their data. We propose measures to make dataspaces scalable and facilitate access to them for a wider variety of actors.
bio
Riccardo Nanni is researcher in data governance at the Digital Commons Lab of Fondazione Bruno Kessler. His background is in international relations, with a PhD obtained in June 2022 at the University of Bologna. His previous research covers diplomatic and public-private relations in the governance of the Internet infrastructure.
Maximising reachability via walk temporalisation
Filippo Brunelli
Tuesday 10 Jan 2023 - Fondazione Bruno Kessler
abstract
In a temporal graph, each edge appears and can be traversed at specific points in time. In such a graph, temporal reachability of one node from another is naturally captured by the existence of a temporal path where edges appear in chronological order. Inspired by the optimisation of bus/metro/tramway schedules in a public transport network, we consider the problem of turning a collection of walks (called trips) in a directed graph into a temporal graph by assigning a starting time to each trip so as to maximise the reachability among pairs of nodes. Each trip represents the trajectory of a vehicle and its edges must be scheduled one right after another. Setting a starting time to the trip thus forces the appearing time of all its edges. We call such a starting time assignment a trip temporalisation. We obtain several results about the complexity of maximising reachability via trip temporalisation. Among them, we show that maximising reachability via trip temporalisation is hard to approximate within a factor $\sqrt{n}/12$ in an $n$-vertex digraph, even if we assume that for each pair of nodes, there exists a trip temporalisation connecting them. On the positive side, we show that there must exist a trip temporalisation connecting a constant fraction of all pairs if we additionally assume symmetry, that is, when the collection of trips to be scheduled is such that, for each trip, there is a symmetric trip visiting the same nodes in reverse order.
bio
Filippo Brunelli received his Master degree in mathematics from the University of Pisa in 2019. He is currently a third year PhD student enrolled at Inria Paris and hosted at IRIF (Insitut de Recherche en Informatique Fondamentale) laboratory in Paris, under the supervision of Laurent Viennot and Pierluigi Crescenzi. His main research interests include temporal graphs and algorithms for transport networks.
The ESA JUICE mission and sub-surface sounding of the Galilean Icy Moons
Elena Donini
Friday 5 May 2023 - Fondazione Bruno Kessler and on-line
abstract
JUpiter Icy moon Explorer (JUICE) mission is the first large European mission to the Jovian system and has onboard ten scientific experiments. JUICE aims to explain how the Solar system works and identify the conditions for planet formation and the emergence of life. FBK is involved in developing the radar sounder instrument, Radar for Icy Moon Exploration (RIME), that images the subsurface by generating radargrams. Radargrams store surface and subsurface geology information, including the possible presence of shallow water pockets on the Icy Moons. The analysis of RIME radargrams acquired on the Galileian Moons will unveil unique surface and subsurface geology information and help better understand the evolution of the Jovian and the solar systems. During the seminar, we will illustrate the JUICE mission focusing on the RIME instrument and the methods for the automatic analysis of radargrams.
bio
Elena Donini is a researcher at the RSDE research unit of FBK. She holds a Ph.D. (cum laude) in Information and Communication Technologies from the University of Trento (Italy) and a master in degree (cum laude) in Telecommunications Engineering from the University of Trento(Italy). Her main research interests are related to the automatic analysis of radar sounder data for Earth observation and planetary exploration.
Charting AI Urbanism
Otello Palmini
Thursday 6 Jun 2024 - Fondazione Bruno Kessler
abstract
The aim of this communication is to tease out some of the key issues concerning the relationship between AI and urbanism. This relationship, which is presented in the academic literature as a new driving force of contemporary urbanism, will be investigated through an interdisciplinary approach that places urban studies and philosophy of technology in dialogue. AI urbanism will be inquired through key turning points in the history of the relationship between technology and the city, namely Modern Urbanism, Cybernetics and the Smart City paradigm. The analysis of these three conceptual sources will provide some insights into AI urbanism, both in terms of criticalities and opportunities. Finally, cybernetics as a model for non-deterministic use of the digital twin will be highlighted.
bio
Bachelor and Master Degree in Philosophy at the University Of Bologna and currently PhD student in Architecture and Urban Planning (IDAUP) at the University of Ferrara. My research is focused on the relationship between Philosophy of Technology and Urban Studies, namely on the epistemological, ethical and environmental implications of Urban AI. Latest Publications: Design culture for Sustainable urban artificial intelligence: Bruno Latour and the search for a different AI urbanism. Ethics and Information Technology, 26(1), 11; (2023) Charting AI urbanism: Conceptual sources and spatial implications of urban artificial intelligence. Discover Artificial Intelligence, 3(1), 15; (2022). Past and future of the connection between project, technology and neocybernetics. AGATHÓN| International Journal of Architecture, Art and Design, 12, 24-35.