Ciclo 38
Finding the most influential nodes in a network is a computationally hard problem with several possible applications in various kinds of network-based problems. Here, we propose an original method, based on network downscaling, that allows a multi-objective evolutionary algorithm (MOEA) to solve the IM problem on a reduced scale network, while preserving the relevant properties of the original network. The downscaled solution is then upscaled to the original network, using a mechanism based on centrality metrics. Our results demonstrate the effectiveness of the proposed method with a more than 10-fold runtime gain compared to the time needed on the original network.
Vedi il poster

Ciclo 37
A counterfactual intervention is a powerful tool which can explain black-box model decisions and enable algorithmic recourse. They are sequences of actions that, if performed by users, can overturn an unfavourable decision made by a machine learning model. Current methods provide interventions without considering the user’s preferences. We propose the first human-in-the-loop approach to perform algorithmic recourse by presenting a novel formalization following the preference elicitation theory.
Vedi il poster

Ciclo 37
Aims at developing a HPC-based data science and learning environment for climate change, with a specific focus on challenges regarding unstructured grid models which provide stronger flexibility to climate modelers. Data-driven/deep learning algorithms, big data concepts and HPC infrastructure are going to be used to create an integrated environment to analyse large volumes of unstructured climate datasets.
Vedi il poster

Ciclo 38
Edge Computing platforms have become increasingly important for enabling smart distributed systems, such as smart cities or buildings and intelligent transportation systems. As a decentralized machine learning approach, Federated learning is a promising solution for training models on edge devices while preserving data privacy. We propose an edge computing platform for federated learning based on Kubernetes and Docker, enabling efficient communication and computation optimization among edge devices. Our proposed platform includes a privacy-preserving mechanism to protect sensitive data during training. We will evaluate the performance of our proposed platform through experiments on a real-world dataset and compare it with existing solutions.
Vedi il poster

Ciclo 38
Class-incremental Novel Class Discovery (class-iNCD), which refers to the problem of discovering novel categories in an unlabelled data set by leveraging a pre-trained model that has been trained on a labelled data set containing disjoint yet related categories. Apart from discovering novel classes, we also aim at preserving the ability of the model to recognize previously seen base categories.
Vedi il poster

National PhD in Artificial Intelligence – Ciclo 37
We introduce Neuro-Symbolic Continual Learning, where a model has to solve a sequence of neuro-symbolic tasks, that is, it has to map sub-symbolic inputs to high-level concepts and compute predictions by reasoning consistently with prior knowledge. Our key observation is that neuro-symbolic tasks, although different, often share concepts whose semantics remain stable over time. Traditional approaches fall short: existing continual strategies ignore knowledge altogether, while stock neuro-symbolic architectures suffer from catastrophic forgetting. We show that leveraging prior knowledge by combining neuro-symbolic architectures with continual strategies does help avoid catastrophic forgetting, but also that doing so can yield models affected by reasoning shortcuts. These undermine the semantics of the acquired concepts, even when detailed prior knowledge is provided upfront and inference is exact, and in turn continual performance. To overcome these issues, we introduce COOL, a COncept-level cOntinual Learning strategy tailored for neuro-symbolic continual problems that acquires high-quality concepts and remembers them over time. Our experiments on three novel benchmarks highlights how COOL attains sustained high performance on neuro-symbolic continual learning tasks in which other strategies fail.
Vedi il poster

Ciclo 37
Human motion modelling aims at understanding and replicating human kinematics. By masking certain body parts or specific motions, it is possible to isolate and focus on the critical aspects of human behaviours. We present a novel approach for motion retargeting, which aims to transfer the movements from a source skeleton to a target one in a different format. We then discuss its potential in many fields, such as healthcare, video surveillance, sports, gaming, and virtual reality.
Il file non è al momento disponibile.

Ciclo 38
Simultaneous speech translation is the process of real-time translation of spoken language into another language, enabling cross-language communication. This technology has become increasingly popular in recent years, with the development of advanced models which, however, still face the main challenge of providing accurate but fast translations. My research aims to fill this gap, by proposing new strategies that leverage the knowledge already acquired by neural models to reduce their latency.
Vedi il poster

Ciclo 37
Technical leverage is the ratio between dependencies and own codes of a software package. We aim to analyze the Python ecosystem by using this metric. We collect a dataset of the top 21,205 Python package versions and used some innovative approaches for its analysis. Our result shows that Python packages ship a lot of other people’s code and tend to keep doing so. On security, the chance of getting a safe package version is actually better than what previous research has reported.
Vedi il poster

Ciclo 36
Developing efficient algorithms for AI verification is a long-standing goal of AI research. To this extent, we propose a framework that combines SMT-based enumeration (an efficient technique in formal verification) and Weighted Model Integration (an extension of Weighted Model Counting) to efficiently perform probabilistic inference on hybrid models. Additionally, we showcase its application in real-world scenarios, such as safety checking of probabilistic programs.
Vedi il poster

Ciclo 37
This work assesses the accuracy and precision of static and dynamic IMU calibration techniques against a gold-standard camera system for the wrist joint. 13 Healthy subjects were instrumented with IMU sensors and retroreflective markers and performed 1) a static N-pose, 2) functional calibration movements, 3) single-plane tasks and 4) multi-joint tasks. Results showed the superiority of a functional calibration method to minimize errors in joint angle estimation for the wrist joint.
Vedi il poster

Ciclo 38
The activity concerns a PhD project in Industrial Innovation aiming to bring robotics for unstructured environments closer to its industrial deployment. Thanks to the introduction of modularity as a fundamental approach to the mechatronic design, it shall be provided an example of robots that can become a permanent asset for a company, deployable in various use-cases and easily repurposed to perform either dynamic motions or high force interaction and replace human operators in dangerous tasks.
Vedi il poster

Ciclo 38
Intrinsically disordered proteins (IDPs) are a class of proteins lacking a well-defined three-dimensional structure under physiological conditions; this peculiarity makes them challenging targets in drug discovery. We propose a fluorescent cell-based technology that will be used in a high-content screening and combined with a fragment-based library to detect modulations in the behavior and concentration of the IDP under study to find effective pharmacophores.
Vedi il poster

Ciclo 38
This work deepens into the Multi-Modal Semantic Mapping problem focusing on the classification and localisation of objects within a map under construction (SLAM) or already built. To further explore this direction, we propose a framework that can autonomously map predefined objects in a known environment using a multi-modal sensor fusion approach (combining RGB and depth data from an RGB-D camera and a lidar). In this way, the robot can understand contextual information along with spatial ones.
Vedi il poster