Researchsummaries
Challenge: Multimodal Semantic Structures
My Solution: Opens up a new direction of multimodel event extraction, to construct a multimodal common semantic space that preserves event semantic structures
My Solution: Opens up a new direction of multimodel event extraction, to construct a multimodal common semantic space that preserves event semantic structures
CLIP-Event: A pioneer to introduce event semantics to vision-language pretraining;
Supports zero-shot vision event extraction for the first time
M2E2: A new task of multimedia event extraction with the first benchmark released
GAIA: The first open-source event-centric multimedia event extraction system
COVID-KG for Drug Repurposing: A knowledge graph extracted from literature, which has been downloaded 2000+ times since 2021
Challenge: Long-Horizon Temporal Dynamics
My Solution: Establishs a new paradigm of Models as Schemas to add predictive power to produce multiple hypotheses, along with structural justifications for participant-specific and attribute-specific connections
My Solution: Establishs a new paradigm of Models as Schemas to add predictive power to produce multiple hypotheses, along with structural justifications for participant-specific and attribute-specific connections
Event Graph Schema: Open doors to graph context for event dynamics
Path Language Model: A new step towards deep semantic understanding of inter-event connections
VidIL: Creates a new paradigm to use language models to understand videos; The event prediction performance without any training videos is higher than models pretrained on 20k+ videos
