Core Research
Multimodal Emotion Understanding
Modeling emotion from speech, vision, and text with depth-aware representations.
Hi, I'm
Zhishuo
Multimodal AI researcher focused on emotion understanding, LLM systems, and practical intelligent products.
About
I am a Ph.D. candidate at the School of Computer Science, Sichuan University. My research centers on multimodal learning, affective computing, agent systems, large models, and robust speech understanding under real-world conditions. I work on multimodal sentiment analysis, cross-modal contrastive optimization, noise-resilient speech recognition, and emotion-aware agent workflows, with an emphasis on interpretability and generalization.
Research Focus
Systems
LLM + MoE Systems
Task-adaptive routing and efficient expert collaboration for better generalization.
Impact
Applied AI Products
Bridging research and deployment through reliable workflows and automation.
Selected Projects
Research
HEME
Hierarchical emotion modeling with adaptive multi-level mixture-of-experts.
Platform
Emotion Agent Stack
End-to-end pipeline for multimodal emotion analysis and conversational AI.
Workflow
AV-RISE
Hierarchical cross-modal denoising for robust audio-visual speech representation under noisy real-world conditions.
Tooling
Paper-to-Product Toolkit
Templates and scripts to accelerate turning academic ideas into usable demos.
News
- 2025π Our paper "AV-RISE: Hierarchical Cross-Modal Denoising for Learning Robust Audio-Visual Speech Representation" is accepted at ACM MM 2025.
- 2024π Published "WinNet: Make Only One Convolutional Layer Effective for Time Series Forecasting" in ICIC 2025.
- 2024π Our paper "AMG-AVSR: Adaptive Modality Guidance for Audio-Visual Speech Recognition via Progressive Feature Enhancement" is accepted at ACML 2024.
Resume
Open CVBlog
Open blogWorkflow
A workflow for research, writing, and review
How I structure literature, experiments, writing, and agent tooling into one continuous loop.
Paper
AV-RISE and robust AV speech
Why cross-modal denoising became central to how I think about noisy real-world speech representation.
Agent
What an emotion agent should optimize for
Not just empathy on the surface, but perception, calibration, clarification, and safe intervention.
Models
Beyond emotion models: conflict, uncertainty, and repair
Why I think the next step is not simply a bigger model, but better handling of conflict, uncertainty, and repair.