Multimodal AI · Large Language Models · Embodied AI
Kehui Tan
I am a master's student at Peking University, interested in multimodal AI, large language models, and embodied AI.
At PKU-HRI, I am advised by Prof. Hong Liu. Previously, I studied Artificial Intelligence at South China Normal University. I welcome research discussions and opportunities for collaboration. Please feel free to reach out at kehui.tt@gmail.com.
Selected publications
Publications →
RA-L 2026
GeoFusion: Geometric Consensus Learning and Anchored Feature Fusion for Multi-View Depth Estimation
Combines geometric consensus-aware cost aggregation with geometrically anchored feature rectification to improve multi-view depth estimation.
JDIQ 2025
ELF: Educational LLM Framework of Improving and Evaluating AI-generated Content for Classroom Teaching
Uses selected teaching examples and prompting to improve classroom dialogue completion and expertise transfer, with automatic and human evaluation.
IJCNN 2025
MiDream: Multi-Domain Retrieval Enhancement with Mixture-of-Experts
Adapts frozen retrieval embeddings with hierarchical domain routing and specialized experts for multi-domain document retrieval.
PRCV 2025
Enhancing Human Trajectory Prediction with Reinforcement Learning from Quantified Human Preferences
Uses a diffusion-based preference score as a reward for trajectory prediction, together with rejection sampling.
IJCNN 2024
REMED: Retrieval-Augmented Medical Document Query Responding with Embedding Fine-Tuning
Improves medical document retrieval through embedding fine-tuning for retrieval-augmented query answering.
Research interests
Multimodal AI
Connecting visual and language representations for richer scene understanding.
Large Language Models
Retrieval-augmented generation, embedding adaptation, and model alignment.
Embodied AI
Geometry-aware perception and intelligent agents that understand and interact with the physical world.