| Attachment | Size |
|---|---|
| 160.16 KB |
Human communication is a little like a dance: participants continuously adjust their behaviors based on their interlocutor’s language, gestures, and facial expressions during social interaction. The actual sophistication of human communication comes to the fore when we try to create computers that can understand and participate, however crudely, in these social interactions. With the advent of new AI and language technologies, it becomes important to understand how computers can understand social interactions, bring it closer to artificial social intelligence. We welcome researchers across disciplines, including NLP, computer vision, robotics and neuroscience, to join our workshop and discuss the science of social intelligence, as well as underlying technical challenges, goals, and ethics of building social intelligence in real-world AI systems. Our workshop will include keynote talks, panel discussions, and paper presentations.
SOCIAL-IQ 2.0 CHALLENGE: Our workshop will host the Social-IQ 2.0 challenge designed to benchmark recent AI technologies in theirs skills to reason about social interactions, which is referred as Artificial Social Intelligence. The Social-IQ 2.0 dataset contains over 1,000 videos, accompanied with 6,000+ multiple-choice questions. The challenge plans to award over $1,200 in total prizes. In order to encourage diverse, innovative approaches, we welcome submissions under one of the following three focuses:
* few-shot focus: composing pre-trained models in a zero- or few-shot way
* fusion and reasoning focus: learning to fuse and integrate information from multimodal features
* representation learning focus: Learning new multimodal representations directly from raw data