Summary of Eliminating the Language Bias For Visual Question Answering with Fine-grained Causal Intervention, by Ying Liu et al.
Eliminating the Language Bias for Visual Question Answering with fine-grained Causal Intervention
by Ying Liu, Ge Bai, Chenji Lu, Shilong Li, Zhang Zhang, Ruifang Liu, Wenbin Guo
First submitted to arxiv on: 14 Oct 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The proposed novel causal intervention training scheme, CIBi, aims to mitigate language bias in Visual Question Answering (VQA) by capturing finer-grained information within a sentence, such as context and keywords. Current approaches only consider coarse-grained perspectives, leading to insufficient capture of language bias. The scheme divides language bias into context bias and keyword bias, using causal intervention and contrastive learning to eliminate context bias and improve multi-modal representation. Additionally, a new question-only branch based on counterfactual generation is designed to distill and eliminate keyword bias. Experimental results demonstrate CIBi’s applicability to various VQA models, achieving competitive performance. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary CIBi is a new way to make sure that Visual Question Answering (VQA) machines are fair and don’t get confused by certain words or phrases in the questions they’re asked. Right now, most VQA systems have a problem where they learn to answer questions based on what words are used, not just what’s being asked. This means they can be biased towards certain topics or ideas. The CIBi system tries to fix this by looking at two types of bias: context and keywords. It uses special techniques to get rid of these biases and create a more fair system. The results show that CIBi works well with different VQA systems and gets good answers. |
Keywords
» Artificial intelligence » Multi modal » Question answering
