Research

AAIG connects five labs working across vision, speech, embodied intelligence, and knowledge-based AI.

Research by Laboratory

Multi-Modal AI

We develop multimodal learning systems that connect visual, language, and diverse real-world signals across computer vision, robotics, efficient LLMs, and applied AI.

Computer Vision & Learning Algorithms: Visual Recognition / Large-scale Models / Meta Learning
Efficient Learning for LLMs: Model Compression / Efficient Learning / System-level Optimization
Robot Learning: Vision-Language-Action / Efficient Robotics / 3D Recognition
Industrial & Medical AI: Battery AI / Fault Detection / Medical AI

Visit MMAI Lab ↗

SAIL

Speech AI & Generative Models

We study speech synthesis, speech language models, and generative approaches for audio, talking-head, and video generation.

Speech Synthesis: Text-to-Speech / Voice Conversion / Neural Vocoder
Speech Language Models: Neural Codec / Speech-to-Speech Translation / Speech Editing
Generative AI: Audio Generation / Talking Head Generation / Video Generation

Visit SAIL ↗

Embodied Intelligence

We build adaptive robots through foundation models, human-robot interaction, and lifelong learning in changing physical environments.

Foundation Models for Robotics: Task Generalization / Cross-environment Adaptation
Human-Robot Interaction: Natural Collaboration / Shared Autonomy
Lifelong Robot Learning: Continuous Skill Acquisition / Interactive Adaptation

Visit HEI Lab ↗

Knowledge-Centered AI

We explore recommendation systems, multimodal understanding, large language models, and federated learning for intelligent knowledge systems.

Recommendation Systems: Personalization / Intelligent Knowledge Systems
Multimodal Understanding: Multimodal Representation / Cross-modal Reasoning
Large Language Models: Knowledge-aware Language Intelligence / Reasoning
Federated Learning: Distributed Learning / Privacy-conscious AI

Visit iKnow Lab ↗

LAMDA Lab

Data-Centric & Trustworthy AI

We develop multimodal, self-supervised, and explainable AI methods for trustworthy analysis of text, images, graphs, biosignals, and real-world data.

Multimodal & Self-Supervised Learning: Representation Learning / Cross-modal Integration
Trustworthy AI: Explainability / Robustness / Safety
Medical & Social Data AI: Biomedical Informatics / Network Analysis
Efficient Language Models: LLM Efficiency / Real-world Deployment

Visit LAMDA Lab ↗