Reproduce “Large Emotional World Model” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Changhao Song, Yazhou Zhang, Hui Gao, Chang Yang, Peng Zhang
Researchers from Tianjin University introduced the Large Emotional World Model (LEWM) and the Emotion-Why-How (EWH) dataset, formalizing human emotion as a dynamic causal factor in world state transitions. LEWM improved video prediction by up to 45.73% on human-centered datasets and achieved absolute gains of 15.78% in emotion recognition accuracy.
Key insights
- Human-centered world models must account for both physical laws and "affective dynamics," where human emotions serve as causal factors driving actions and shaping future environmental states.
- Explicitly modeling future emotional states before predicting world states allows for more accurate and contextually informed predictions in human-centered environments.
- Emotional dynamics provide complementary cues that significantly improve multimodal state prediction, especially for visual and audio modalities, without compromising general reasoning capabilities.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.