DeVLBert: Learning Deconfounded Visio-Linguistic Representations
This addresses generalization issues in visio-linguistic models for researchers and practitioners, but it is incremental as it builds on existing Bert-style methods with causal adjustments.
The paper tackles the problem of out-of-domain visio-linguistic pretraining by addressing spurious correlations in data that hurt generalization, proposing DeVLBert to mitigate dataset biases and showing effectiveness in boosting generalization on tasks like Image Retrieval and Visual Question Answering.
In this paper, we propose to investigate the problem of out-of-domain visio-linguistic pretraining, where the pretraining data distribution differs from that of downstream data on which the pretrained model will be fine-tuned. Existing methods for this problem are purely likelihood-based, leading to the spurious correlations and hurt the generalization ability when transferred to out-of-domain downstream tasks. By spurious correlation, we mean that the conditional probability of one token (object or word) given another one can be high (due to the dataset biases) without robust (causal) relationships between them. To mitigate such dataset biases, we propose a Deconfounded Visio-Linguistic Bert framework, abbreviated as DeVLBert, to perform intervention-based learning. We borrow the idea of the backdoor adjustment from the research field of causality and propose several neural-network based architectures for Bert-style out-of-domain pretraining. The quantitative results on three downstream tasks, Image Retrieval (IR), Zero-shot IR, and Visual Question Answering, show the effectiveness of DeVLBert by boosting generalization ability.