Summary of Don’t Buy It! Reassessing the Ad Understanding Abilities Of Contrastive Multimodal Models, by A. Bavaresco et al.
Don’t Buy it! Reassessing the Ad Understanding Abilities of Contrastive Multimodal Models
by A. Bavaresco, A. Testoni, R. Fernández
First submitted to arxiv on: 31 May 2024
Categories
- Main: Computation and Language (cs.CL)
- 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 paper investigates the capabilities of contrastive vision-and-language models (VLMs) in understanding image-based advertisements. It’s found that these models can solve an ad-explanation retrieval task by exploiting grounding heuristics, which are biases that humans tend to use when interpreting multimodal stimuli. To control for this confound, the authors introduce TRADE, a new evaluation test set with adversarial grounded explanations that “fool” four different contrastive VLMs. The study highlights the need for an improved operationalisation of automatic ad understanding that truly evaluates VLMs’ multimodal reasoning abilities. |
Low | GrooveSquid.com (original content) | Low Difficulty Summary Automatic ads are complex and often use unusual visual elements and figurative language. Researchers have used special computer models to understand these ads, and they’ve had great success on a task called ad-explanation retrieval. However, this study shows that the models can solve the task by using biases we humans tend to apply when interpreting multimodal stimuli. To test how well the models really work, the authors created a new set of explanations for ads that are designed to trick the models. The results show that four different models were fooled by these fake explanations. |
Keywords
» Artificial intelligence » Grounding