Explaining Multi-modal Large Language Models by Analyzing their Vision Perception

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Giulivi, Loris, Boracchi, Giacomo
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866914814639472640
author Giulivi, Loris
Boracchi, Giacomo
author_facet Giulivi, Loris
Boracchi, Giacomo
contents Multi-modal Large Language Models (MLLMs) have demonstrated remarkable capabilities in understanding and generating content across various modalities, such as images and text. However, their interpretability remains a challenge, hindering their adoption in critical applications. This research proposes a novel approach to enhance the interpretability of MLLMs by focusing on the image embedding component. We combine an open-world localization model with a MLLM, thus creating a new architecture able to simultaneously produce text and object localization outputs from the same vision embedding. The proposed architecture greatly promotes interpretability, enabling us to design a novel saliency map to explain any output token, to identify model hallucinations, and to assess model biases through semantic adversarial perturbations.
format Preprint
id arxiv_https___arxiv_org_abs_2405_14612
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Explaining Multi-modal Large Language Models by Analyzing their Vision Perception
Giulivi, Loris
Boracchi, Giacomo
Computer Vision and Pattern Recognition
Artificial Intelligence
Multi-modal Large Language Models (MLLMs) have demonstrated remarkable capabilities in understanding and generating content across various modalities, such as images and text. However, their interpretability remains a challenge, hindering their adoption in critical applications. This research proposes a novel approach to enhance the interpretability of MLLMs by focusing on the image embedding component. We combine an open-world localization model with a MLLM, thus creating a new architecture able to simultaneously produce text and object localization outputs from the same vision embedding. The proposed architecture greatly promotes interpretability, enabling us to design a novel saliency map to explain any output token, to identify model hallucinations, and to assess model biases through semantic adversarial perturbations.
title Explaining Multi-modal Large Language Models by Analyzing their Vision Perception
topic Computer Vision and Pattern Recognition
Artificial Intelligence
url https://arxiv.org/abs/2405.14612