Dipartimento di Ingegneria informatica, automatica e gestionale

Carissime/i,

vi invito a partecipare al seguente seminario, in cui Edoardo Barba illustrerà la propria attività di ricerca, in ottemperanza ai requisiti previsti dalla procedura valutativa per n.3 posti di Ricercatore a Tempo Determinato tipologia A - SC 09/H1 SSD ING-INF/05 - Dipartimento di Ingegneria Informatica Automatica e Gestionale "A. Ruberti", 2023RTDAPNRR001, pubblicato sulla Gazzetta Ufficiale N. 5 del 20.01.2023. Il seminario sarà fruibile anche in modalità telematica via Zoom all'indirizzo: https://uniroma1.zoom.us/j/97615093680?pwd=R3U0bDM3aW1qVHM3TGsxYkd0OTc3Zz09

Grazie,
Roberto

Title: Neuro-symbolic approaches: towards safe, interpretable and explainable NLP

Abstract

Large language models (LLMs) have become central to virtually every natural language processing (NLP) task. They are utilized in a wide range of everyday applications, including chatbots and translation systems, by millions of users worldwide. Big-tech companies and major research groups are investing significant resources in creating more powerful LLMs by using large amounts of data and increasing the number of parameters. However, this blind pursuit of more powerful models has come at a cost. This seminar will discuss the environmental impact, biases, and limitations of large LLMs, as well as the difficulty of interacting with them and explaining their behaviour. Most importantly, we will explore a possible solution to these challenges by introducing a novel paradigm for Neuro-symbolic approaches. We will present the main characteristics of this paradigm and explain why it tackles most of the challenges previously introduced. Ultimately, this seminar will argue that the future of NLP lies in the development of more interpretable and transparent models, capable of taking into account human input and feedback, and that can be used to solve real-world problems in a responsible and ethical manner.


Bio: Edoardo Barba is a postdoc researcher at the Department of Computer, Control and Management Engineering (DIAG), Sapienza University of Rome, working in the area of Natural Language Processing in the Sapienza NLP Group led by prof. Roberto Navigli. His interests are in lexical semantics, sentence-level semantics and retrieval-augmented neuro-symbolic generative approaches. His research has been published in conferences such as the ACL, EMNLP, IJCAI and AAAI.