Research on Multi-Modal Large Language Models and their application for the verification and validation of identity documents

Consultable a partir de

2029-10-01

Date

2024

Authors

Otazu Redín, Judit

Publisher

Acceso embargado / Sarbidea bahitua dago
Trabajo Fin de Máster / Master Amaierako Lana

Project identifier

Abstract

In response to the exponential growth and increasing adoption of Multi-Modal Large Language Models, this project aims to explore their application in a critical field: the verification and validation of identity documents. These models, which effectively integrate image, text, video, and audio processing, are proposed as potential improvements over traditional systems specialized in specific tasks. The research will compare the effectiveness of MM-LLMs against dedicated models, including both commercial and open-source solutions, in key tasks such as classification, image quality, fraud detection, OCR (Optical Character Recognition) and Entity Mapping. Additionally, the explainability of these multimodal models will be analyzed, offering a transparent alternative to the opacity of the ’black box’ typically associated with artificial intelligence. The study also recognizes and addresses the challenges that arise from the substantial hardware demands and potential latency issues inherent in these advanced systems.

Description

Keywords

Large Language Model, Multi-Modal Large Language Model, Natural Language Processing, Computer Vision, Transformers, Embeddings, CNN, OCR (Optical Character Recognition), Document Authenticity, Anti-spoofing, Prompting

Department

Faculty/School

Escuela Técnica Superior de Ingeniería Industrial, Informática y de Telecomunicación / Industria, Informatika eta Telekomunikazio Ingeniaritzako Goi Mailako Eskola Teknikoa

Degree

Máster Universitario en Ingeniería Informática por la Universidad Pública de Navarra, Nafarroako Unibertsitate Publikoko Unibertsitate Masterra Informatika Ingeniaritzan

Doctorate program

item.page.cita

item.page.rights

Los documentos de Academica-e están protegidos por derechos de autor con todos los derechos reservados, a no ser que se indique lo contrario.