Behind the Scene: Revealing the Secrets of Pre-trained Vision-and-Language Models

Abstract

Recent Transformer-based large-scale pre-trained models have revolutionized vision-and-language (V+L) research. Models such as ViLBERT, LXMERT and UNITER have significantly lifted state of the art across a wide range of V+L benchmarks with joint image-text pre-training. However, little is known about the inner mechanisms that destine their impressive success. To reveal the secrets behind the scene of these powerful models, we present VALUE (Vision-And-Language Understanding Evaluation), a set of meticulously designed probing tasks (e.g., Visual Coreference Resolution, Visual Relation Detection, Linguistic Probing Tasks) generalizable to standard pre-trained V+L models, aiming to decipher the inner workings of multimodal pre-training (e.g., the implicit knowledge garnered in individual attention heads, the inherent cross-modal alignment learned through contextualized multimodal embeddings). Through extensive analysis of each archetypal model architecture via these probing tasks, our key observations are: (i) Pre-trained models exhibit a propensity for attending over text rather than images during inference. (ii) There exists a subset of attention heads that are tailored for capturing cross-modal interactions. (iii) Learned attention matrix in pre-trained models demonstrates patterns coherent with the latent alignment between image regions and textual words. (iv) Plotted attention patterns reveal visually-interpretable relations among image regions. (v) Pure linguistic knowledge is also effectively encoded in the attention heads. These are valuable insights serving to guide future work towards designing better model architecture and objectives for multimodal pre-training.

arXiv commentAccepted by ECCV 2020 as Spotlight

References in corpus (15)

• VisualBERT: A Simple and Performant Baseline for Vision and Language
• VQA: Visual Question Answering
• SentEval: An Evaluation Toolkit for Universal Sentence Representations
• Large-Scale Adversarial Training for Vision-and-Language Representation Learning
• Visual Genome: Connecting Language and Vision Using Crowdsourced Dense Image Annotations
• Learning Video Representations using Contrastive Bidirectional Transformer
• UNITER: UNiversal Image-TExt Representation Learning
• Visual Entailment: A Novel Task for Fine-Grained Image Understanding
• What do you learn from context? Probing for sentence structure in contextualized word representations
• Generation and Comprehension of Unambiguous Object Descriptions
• Unicoder-VL: A Universal Encoder for Vision and Language by Cross-modal Pre-training
• Language Models as Knowledge Bases?
• HERO: Hierarchical Encoder for Video+Language Omni-representation Pre-training
• Probing Neural Network Comprehension of Natural Language Arguments
• Evaluating Commonsense in Pre-trained Language Models

Cited by in corpus (6)

• Large-Scale Adversarial Training for Vision-and-Language Representation Learning
• UNITER: UNiversal Image-TExt Representation Learning
• Trends in Integration of Vision and Language Research: A Survey of Tasks, Datasets, and Methods
• A Closer Look at the Robustness of Vision-and-Language Pre-trained Models
• A Comparison of Pre-trained Vision-and-Language Models for Multimodal Representation Learning across Medical Images and Reports
• Exceeding the Limits of Visual-Linguistic Multi-Task Learning