ATTENTION/WARNING - NE PAS DÉPOSER ICI/DO NOT SUBMIT HERE

Ceci est la version de TEST de DIAL.mem. Veuillez ne pas soumettre votre mémoire sur ce site mais bien à l'URL suivante: 'https://thesis.dial.uclouvain.be'.
This is the TEST version of DIAL.mem. Please use the following URL to submit your master thesis: 'https://thesis.dial.uclouvain.be'.
 

Memory Augmented Generator : a new approach for question-answering

Langlois, Quentin
(2022)

Files

Langlois_19281700_2022.pdf
  • Closed access
  • Adobe PDF
  • 6.56 MB

Details

Supervisors
Dupont, Pierre
Faculty
Ecole polytechnique de Louvain
Degree label
Master [120] en sciences informatiques, à finalité spécialisée
Abstract
Nowadays, Question-Answering (Q&A), a specific domain of Natural Language Processing (NLP), is widely used in real-world applications such as chatbots, Voice Assistants, etc. However, Transformers, a Deep Learning architecture achieving most of the State Of The Art (SOTA) results in various NLP tasks, are not used in real-world Q&A applications due to some limitations. The main limitation comes from design choices : they improve performances, but allow models to only handle a restricted number of words. This Master Thesis introduces Memory Augmented Generator (MAG), a new way to pass data through the classical models, to allow them to handle as many paragraphs as allowed by the hardware (e.g. GPU memory). The objectives are to demonstrate the performances of the approach in a real-world context, to compare it to popular systems (Google Assistant and Siri) and to show its adaptability to any source of textual knowledge.