MODAL@UNINA

G.A.N.D.A.L.F – Gan Approaches for Non-iiD Aiding Learning in Federations

PRIN 2022 – Principal Investigator: Prof. Francesco Piccialli

Abstract:  Federated Learning (FL) is currently one of the hottest research topics in the machine learning community. Introduced by Google in 2016, the hunger for data sharing and the privacy-security tandem will favor its rise. What if we wanted to collaborate, but without sharing data?
Let’s think about the medical field and the experience of the Covid-19 pandemic. The lesson we have learned is that in order to reach effective diagnoses and therapies in a short time, especially when everyone has limited knowledge, it is essential to have a collaborative learning model. However, sharing data is a problem for many reasons (practicals, competitive advantages (business data) or laws (health data)). While classical Deep Learning (DL) requires a huge amount of centralized data, FL allows (i) users to train an algorithm across multiple decentralized databases, (ii) a fruitful collaboration with private data.
Notably, the performance of FL models is drastically reduced when data are not independent and identically distributed (non-iid). In addition, in situations where one of the members of the federation can maliciously attack others, reducing the data integrity can also limit the global performance.
The GANDALF project aims to address one of the most FL challenges, the non-IID problem, in light of the wider scenario of the Edge Artificial Intelligence (Edge-AI). Edge-AI is a modern way of doing Machine Learning allowed by computationally more efficient edge devices (i.e. Google Coral and/or Nvidia Jetson). The distribution of classes across devices must be as close as possible for FL to perform well, especially on the Edge. The output degrades when local dataset distributions are extremely inconsistent and non-IID.
Moreover, GANDALF will introduce a suitable Blockchain-based methodology to mitigate the incidence of security-related attacks on the data concerning the methodologies used for the non-IID problem.

———————————————

 D.I.R.E.C.T.I.O.N.S. – Deep learning aIded foReshock deteCTIOn Of iNduced mainShocks

PRIN PNRR 2022 – Research Unit Coordinator: Prof. Francesco Piccialli

Abstract: Worldwide demand for clean and sustainable energy is forcing the exploitation of new sources. Indeed, the increment of energy production from renewable sources is a key point and requirement in the European recovery and resilience plan (PNRR). Among the most exploited sources, energy from geothermal areas has become one of the most used and sustainable. However, an unwanted consequence of field operations aimed at exploiting new energy sources involving earth’s interior is induced seismicity. In fact, fluid injection or extraction can perturb the local stress field producing new fractures – for example during fracking operations – or activating existing faults due to, for example,
normal stress reduction by pore fluid pressure increase. Almost all documented induced earthquakes have limited magnitude value, particularly those related to wastewater injection, they generally occur at relatively shallow depths possibly producing damage or annoyance to inhabitants close to the facility that can lead to the halting of the field operations with high economic losses. Although the deterministic prediction of relevant earthquakes is feasible in practice, several efforts have been made to identify the preparatory phase of an impending earthquake. To this aim, seismologists are profusing significant efforts to detect foreshocks that are the most obvious premonitory observable. Anyway, current techniques are based on a high degree of subjectivity in the proper selection of both space and time windows in which foreshocks should be searched.
The aim of the present project is to provide a new approach for the detection of foreshocks that is based on the research of patterns in the evolution of the seismicity recorded during field operations. Specifically, the project will benefit from the recent advances gained by artificial intelligence and, in particular, by deep learning (DL) algorithms.

———————————————

C.L.A.I.M. – Artificial Intelligence for Competences and Learning

Erasmus+, KA220-VET, EU, 2023 – 2025 – Principal Investigator: Prof. Francesco Piccialli

Abstract: European companies are undergoing transformations in internationalization, digital transition, and sustainability, reshaping their internal structures, operational practices, organizational configurations, and business models. New artificial intelligence technologies play a pivotal role in these processes, combining human-centered approaches with sophisticated methods for qualitative and quantitative analysis of needs and potentials. The C.L.A.I.M. project aims to equip SMEs with innovative technological tools for personalized analysis of training and professional needs, avoiding generic training approaches. Instead, it focuses on a competence-based model, recognizing individual skills and know-how, driving excellence and allowing European SMEs to stand out.

———————————————

Deep-Learning-aided GPC-IR fingerprinting of complex polyolefin mixtures

2022 CALL FOR RESEARCH PROPOSALS, RESEARCH PROPOSAL TO THE DUTCH POLYMER INSTITUTE

Principal Investigator and Applicant: Prof. Francesco Piccialli

Abstract: A sustainable society needs plastics, and also practical ways for recycling post-consumer plastic wastes. This project addresses the latter question for polyolefins (whose share of the plastic market already exceeds 50 wt.-% and is predicted to grow further) with an interdisciplinary approach. The general idea is to facilitate the mechanical recycling of polyolefin waste streams, previously separated from other plastics with existing methods, by implementing a rapid fine sorting instrument which integrates high-end characterization techniques and a properly designed Artificial Intelligence algorithm, trained on a large archive of molecular fingerprints for commercial grades and able to recognize said fingerprints in complex mixtures.

———————————————

ELIXIR x NextGenerationIT: Consolidamento dell’Infrastruttura Italiana per i Dati Omici e la Bioinformatica (ElixirxNexGentIT)

Avviso n. 3264 del 28/12/2021 “Rafforzamento e  creazione di IR nell’ambito del Piano Nazionale di Ripresa e Resilienza (PNRR)

Research Unit Coordinator: Prof. Francesco Piccialli

Abstract: The aim of the project “ELIXIR x NextGenerationIT: consolidation of the Italian Infrastructure for Omics Data and Bioinformatics” (ELIXIRxNextGenIT) is to consolidate the capacity of the Italian node of the European Research Infrastructure ELIXIR to develop, provide and support reliable services for the Life Science research community and beyond. ELIXIRxNextGenIT will endow the country with a state-of-the-art Bioinformatics and integrative Omics infrastructure, providing access to the most advanced platforms for high-throughput generation and analysis of genomic, proteomic and metabolomic data.

———————————————

4.I.  – mixed reality, machine learning, gamification and educational for Industry

M.I.S.E. Prog n. F/190130/02/X44 – a valere sull’Asse 1, azione 1.1.3. del Programma Operativo Nazionale «Imprese e Competitività» 2014-2020 FESR

Principal Investigator: Prof. Francesco Piccialli

Abstract: Il progetto 4I, diretto al miglioramento dei processi esistenti, intende proporre alle industrie un nuovo modello di condivisione della conoscenza, di supervisione e manutenzione dei processi e delle attrezzature e di formazione del personale e QA, Quality Assurance, sfruttando le nuove tecnologie di mixed reality, il machine learning, l’educational e la gamification.

———————————————

BIOCHIP – Intelligent biosensors based on chimeric proteins 

Programma per il Finanziamento della Ricerca di Ateneo (FRA 2020)

Research Unit Coordinator: Prof. Salvatore Cuomo

Abstract: The project “Intelligent biosensors based on chimeric proteins” (BIOCHIP) will synergically integrate principles and potentialities derived from biochemistry, chemistry, and informatics, to create intelligent biosensors endowed with high sensitivity, specificity, reliability, and ability to work in a wide range of matrices.
A variety of chimeric proteins endowed with both the adhesive properties of a selfassembling amyloid moiety and the recognition ability of specific proteins will be designed for monitoring model analytes of interest in the diagnostic and food fields.

———————————————

C.E.T.R.A.  – Cultural Equipment with Transmedial Recommendation Analytics

POR FESR CAMPANIA 2014/2020- O.S. 1.1– Avviso Pubblico per il Sostegno alle Imprese nella realizzazione di studi di fattibilità (Fase 1) e Progetti di Trasferimento Tecnologico (Fase 2) coerenti con la RIS 3.

Principal Investigator: Prof. Francesco Piccialli

Abstract: Il progetto porta avanti una combinazione originale di IoT, machine learning, recommendation, content adaptivity e Big Data Analytics secondo il paradigma della transmedialità. Esso si propone di adottare metodologie e strumentazione analitica per la Data Science, applicate alla combinazione e integrazione delle conoscenze ottenibili dalle varie fonti eterogenee di rilevazione e monitoraggio. L’obiettivo è lo sviluppo di paradigmi innovativi per processi turistici – culturali in senso esteso, con strumenti applicabili tanto in siti fissi /permanenti (musei, siti culturali / turistici attrezzati), quanto in attività episodiche /itineranti di varia natura (stand espositivi, fiere, mostre, convegni, eventi sociali / culturali).

———————————————

Modelli matematici per l’efficientamento energetico degli edifici

PON Ricerca ed Innovazione 2014-2020

Abstract: Temi di grande attualità come l’ecosostenibilità, il risparmio energetico, il miglioramento ambientale, l’efficientamento energetico, non possono prescindere dai problemi di isolamento termico. Per questo motivo quest’area di ricerca è di cruciale interesse e coinvolge settori molto diversi che vanno dall’ingegneria civile (che si occupa di progettare edifici in maniera sempre più efficiente), alla fisica e alla chimica (che si occupano della ricerca di materiali innovativi con proprietà e prestazioni sempre più elevate), fino ad arrivare alla matematica (che si occupa dello studio delle equazioni differenziali alle derivate parziali che modellizzano conduzione, diffusione e irraggiamento del calore). Quest’ultimo aspetto costituisce il filone principale della ricerca proposta. Studiosi del Dipartimento di Matematica e Applicazioni della Federico II lavorano da alcuni anni con colleghi dell’Università di Pisa e dell’Università della Savoia su problematiche riguardanti l’ottimizzazione della disposizione di materiali isolanti.

———————————————

CUP-i-One – CUP in un Click

(In Collaborazione con il CINI)

POR FESR CAMPANIA 2014/2020- O.S. 1.1– Avviso Pubblico per il Sostegno alle Imprese nella realizzazione di studi di fattibilità (Fase 1) e Progetti di Trasferimento Tecnologico (Fase 2) coerenti con la RIS 3.

Research Unit Coordinator: Prof. Francesco Piccialli

Abstract: Il progetto si pone l’obiettivo di illustrare l’attuale livello funzionale e tecnologico della piattaforma Cup (Centro Unificato di Prenotazione) e le relative azioni di sviluppo sperimentale e ricerca industriale tese a rendere il sistema innovativo dal punto di vista tecnologico e ricco di nuove funzionalità. Le azioni intraprese saranno finalizzate ad assicurare la piena applicazione di quanto indicato nella eHealth Information Strategy Nazionale, ritenuto basilare per la raccolta delle informazioni in ambito Sanità Elettronica e per il sistema socio-sanitario, a garantire quanto indicato nelle Linee Guida nazionali ai fini del rispetto dei LEA e a rispettare quanto definito nel GDPR in materia di sicurezza e trattamento dei dati sensibili. Partendo dall’attuale core infrastrutturale, Il nuovo modello applicativo proposto sarà in grado di utilizzare meccanismi di System Integration, Interoperabilità e cooperazione applicativa avanzati e riconosciuti come standard de facto e di abbinare soluzioni innovative circa l’analisi puntuale ed approfondita dei flussi informativi raccolti dalla piattaforma, proponendo KPI (Key Performance Indicatori) avanzati per lo studio degli scenari lavorativi.