Summary of Just Kiddin: Knowledge Infusion and Distillation For Detection Of Indecent Memes, by Rahul Garg et al.
Just KIDDIN: Knowledge Infusion and Distillation for Detection of INdecent Memes
by Rahul Garg, Trilok Padhi, Hemang Jain, Ugur Kursuncu, Ponnurangam Kumaraguru
First submitted to arxiv on: 19 Nov 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Computer Vision and Pattern Recognition (cs.CV)
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 This paper presents a novel framework that combines Knowledge Distillation (KD) from Large Visual Language Models (LVLMs) with knowledge infusion to improve toxicity detection in hateful memes. The proposed approach extracts sub-knowledge graphs from ConceptNet, a large-scale commonsense Knowledge Graph (KG), and infuses them into a compact VLM framework. This enables the model to reason about toxic phrases in captions and memes, as well as visual concepts in memes. Experimental results on two hate speech benchmark datasets demonstrate superior performance over state-of-the-art baselines across AU-ROC, F1, and Recall metrics, with improvements of 1.1%, 7%, and 35%, respectively. The paper highlights the significance of learning from both explicit (KG) and implicit (LVLMs) contextual cues incorporated through a hybrid neurosymbolic approach. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about making computers better at identifying harmful content on the internet, like mean memes. Right now, computers struggle to understand the connections between text and images, which makes it hard for them to spot toxic material. The researchers propose a new way to train computers using a combination of big language models and knowledge from the internet. This helps computers learn about the relationships between words and images, making them better at detecting mean memes. In tests on two different datasets, this approach outperformed other methods by a lot. This is important because accurate computer systems can help create safer online environments. |
Keywords
» Artificial intelligence » Knowledge distillation » Knowledge graph » Recall
