Summary of Genearl: a Training-free Generative Framework For Multimodal Event Argument Role Labeling, by Hritik Bansal et al.
GenEARL: A Training-Free Generative Framework for Multimodal Event Argument Role Labeling
by Hritik Bansal, Po-Nien Kung, P. Jeffrey Brantingham, Kai-Wei Chang, Nanyun Peng
First submitted to arxiv on: 7 Apr 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 GenEARL framework addresses the challenge of multimodal event argument role labeling (EARL) without relying on high-quality event-annotated training data. By harnessing modern generative models, GenEARL understands event task descriptions given image contexts and performs EARL tasks. The framework consists of two stages: a frozen vision-language model (VLM) learns the semantics of event argument roles and generates event-centric object descriptions based on the image, followed by a frozen large language model (LLM) prompted with the generated object descriptions using a predefined template for EARL. GenEARL outperforms the contrastive pretraining (CLIP) baseline by 9.4% and 14.2% accuracy for zero-shot EARL on the M2E2 and SwiG datasets, respectively. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary GenEARL is a new way to understand images and events without needing lots of labeled training data. It uses two types of models: one that looks at pictures and understands what’s happening (VLM), and another that reads text and generates descriptions (LLM). These models work together to figure out the roles of objects in an image based on the event described. This is important because it allows us to understand images better without needing lots of labeled data. |
Keywords
» Artificial intelligence » Language model » Large language model » Pretraining » Semantics » Zero shot
