Summary of Bovila: Bootstrapping Video-language Alignment Via Llm-based Self-questioning and Answering, by Jin Chen et al.
BoViLA: Bootstrapping Video-Language Alignment via LLM-Based Self-Questioning and Answering
by Jin Chen, Kaijing Ma, Haojian Huang, Jiayu Shen, Han Fang, Xianghao Zang, Chao Ban, Zhongjiang He, Hao Sun, Yanmei Kang
First submitted to arxiv on: 17 Sep 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 BoViLA framework is a self-training approach that enhances question samples during training by leveraging large language model (LLM) based self-questioning and answering. This helps the model better utilize video information and internal LLM knowledge to improve modality alignment in video question answering (VideoQA) tasks. The framework also includes an Evidential Deep Learning (EDL) component that estimates uncertainty and filters out low-quality self-generated questions by evaluating modality alignment within context. BoViLA outperforms several state-of-the-art methods on five strong VideoQA benchmarks, demonstrating its effectiveness and generality. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary A new way to help machines understand videos better is presented in this paper. The problem is that it’s hard and expensive to create labeled data for training video-understanding models. To make things easier, the authors propose a self-training approach called BoViLA. This framework uses large language models to generate new questions and answers based on existing ones, which helps the model learn more about videos. The authors also developed a way to filter out bad generated questions by checking how well they align with the video. The results show that this approach works well and can be used for various video-understanding tasks. |
Keywords
» Artificial intelligence » Alignment » Deep learning » Large language model » Question answering » Self training
