Suppressing Biased Samples for Robust VQA

Ouyang, Ninglin; Huang, Qingbao; Li, Pijian; Cai, Yi; Liu, Bin; Leung, Ho-fung; Li, Qing

doi:10.1109/tmm.2021.3097502

Cited by 26 publications

(1 citation statement)

References 40 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Dataset Bias is another problem preventing the vision language models from learning better representations. The dataset bias [55], for example only woman wearing dress in this dataset, leads the model to learn biased features. In such a case, the model will simply remember the correlation between woman and dress instead of looking at the image for an answer.…”

Section: G Pretrained Vision Language Model Therapy With Neur-logxmentioning

confidence: 99%

Neural Logic Vision Language Explainer

Yang,

Liu,

Lin

2024

IEEE Trans. Multimedia

View full text Add to dashboard Cite

If we compare how humans reason and how deep models reason, humans reason in a symbolic manner with a formal language called logic, while most deep models reason in black-box. A natural question to ask is "Do the trained deep models reason similar as humans?" or "Can we explain the reasoning of deep models in the language of logic?" . In this work, we present NeurLogX to explain the reasoning process of deep vision language models in the language of logic. Given a trained vision language model, our method starts by generating reasoning facts through augmenting the input data. We then develop a differentiable inductive logic programming framework to learn interpretable logic rules from the facts. We show our results on various popular vision language models. Interestingly, we observe that almost all of the tested models can reason logically.

show abstract

Section: G Pretrained Vision Language Model Therapy With Neur-logxmentioning

confidence: 99%