EgoLink Challenge, ACM Multimedia 2026
Developed SAGE, a multimodal agent system for linking and reasoning over egocentric interaction evidence.
Role: Lead contributor to SAGE
Certificate image to be added
Recognition & Service
Competition results, honors, awards, and academic service by Zebang Cheng.
Developed SAGE, a multimodal agent system for linking and reasoning over egocentric interaction evidence.
Role: Lead contributor to SAGE
Certificate image to be added
A multimodal affective-computing challenge focused on understanding emotional content in visual art.
Role: Team member
Certificate image to be added
A challenge result in audio-visual generation and multimodal perception under the ACM Multimedia 2026 competition program.
Role: Team member
Certificate image to be added
Improved Emotion-LLaMA with Conv-Attention for robust multimodal emotion recognition under noisy conditions.
Role: Core contributor
Certificate image to be added
Used multimodal language-model reasoning to support open-vocabulary emotion recognition beyond a fixed label set.
Role: Core contributor
Certificate image to be added
Built a multimodal system for extracting emotion-cause pairs from conversations using multimodal language models.
Role: Co-first author and core contributor
Certificate image to be added
Introduced Expression MAE, a semi-supervised approach for learning robust multimodal emotion representations.
Role: First author and core contributor
Certificate image to be added