SynBA: A contextualized Synonim-Based adversarial Attack for text classification

With the advent of high-performance computing devices, deep neural networks have gained a lot of popularity in solving many Natural Language Processing tasks. However, they are also vulnerable to adversarial attacks, which are able to modify the input text in order to mislead the target model. Adversarial attacks are a serious threat to the security of deep neural networks, and they can be used to craft adversarial examples that steer the model towards a wrong decision. In this dissertation, we propose SynBA, a novel contextualized synonym-based adversarial attack for text classification. SynBA is based on the idea of replacing words in the input text with their synonyms, which are selected according to the context of the sentence. We show that SynBA successfully generates adversarial examples that are able to fool the target model with a high success rate. We demonstrate three advantages of this proposed approach: (1) effective - it outperforms state-of-the-art attacks by semantic similarity and perturbation rate, (2) utility-preserving - it preserves semantic content, grammaticality, and correct types classified by humans, and (3) efficient - it performs attacks faster than other methods.

MongoDB: analisi e prototipazione su applicazioni di social business intelligence

L’ultimo decennio ha visto un radicale cambiamento del mercato informatico, con la nascita di un numero sempre maggiore di applicazioni rivolte all’interazione tra utenti. In particolar modo, l’avvento dei social network ha incrementato notevolmente le possibilità di creare e condividere contenuti sul web, generando volumi di dati sempre maggiori, nell’ordine di petabyte e superiori. La gestione di tali quantità di dati ha portato alla nascita di soluzioni non relazionali appositamente progettate, dette NoSQL. Lo scopo di questo documento è quello di illustrare come i sistemi NoSQL, nello specifico caso di MongoDB, cerchino di sopperire alle difficoltà d’utilizzo dei database relazionali in un contesto largamente distribuito. Effettuata l'analisi delle principali funzionalità messe a disposizione da MongoDB, si illustreranno le caratteristiche di un prototipo di applicazione appositamente progettato che sfrutti una capacità peculiare di MongoDB quale la ricerca full-text. In ultima analisi si fornirà uno studio delle prestazioni di tale soluzione in un ambiente basato su cluster, evidenziandone il guadagno prestazionale.

Traduzione per il web nelle istituzioni europee. Proposta di localizzazione di una sezione di sito della dg azione per il clima

The aim of this dissertation is to provide an adequate translation from English into Italian of a section of the European Commission's site, concerning an environmental policy tool whose aim is to reduce the EU greenhouse gas emissions, the Emissions Trading System. The main reason behind this choice was the intention to combine a personal interest in the domain of sustainability development with the desire to delve deeper into the knowledge of the different aspects involved in the localisation process. I also had the possibility to combine these two with my interest in the universe of the European Union. I therefore worked on the particular language of this supranational organisation and for this reason I had the opportunity to experience a very stimulating work placement at the Directorate-General for Translation in Brussels. However, the choice of the text was personal and the translation is not intended for publication. The work is divided into six chapters. In the first chapter the text is contextualised within the framework of the EU, and its legislation on multilingualism. This has consequences on the languages that are used by the drafters of the official documents and on the languages used by translators. The text originates from those documents, but it needs to be adapted to different receivers. The second chapter investigates the process of website localisation. The third chapter offers an analysis of the source text and of the prospective target text. In the fourth chapter the resources created and used for the translation of the text are described. A comparison is made between the resources of the translation service of the European Commission and the ones created specifically for this project: a translation memory, exploited through the use of a CAT tool, and two corpora. The fifth chapter contains the actual translation, side-by-side with the source text, while the sixth one provides a comment on the translation strategies.