EMNLPMain Conference paper
CRAM: Centroid-Routing and Adaptive MoE for Multimodal Continual Instruction Tuning
Conference on Empirical Methods in Natural Language Processing. EMNLP 2026
Jun-Tao Tang
Undergraduate researcher · Nanjing University
Jun-Tao Tang
I am an undergraduate student in the School of Computer Science at Nanjing University. Since 2025, I have been a research intern in the LAMDA Group under the supervision of Pr. Da-Wei Zhou. My work explores how intelligent systems learn, adapt, and generate across modalities.
Current interests
World ModelContinual Post-TrainingGenerative Vision (Generative AI)
Featured paper
CRAM, in motion.
Paper walkthrough
PreprintPreprint
ProtoAda: Prototype-Guided Adaptive Adapter Expansion and Geometric Consolidation for Multimodal Continual Instruction Tuning
Preprint · arXiv:2606.02576
A compact view of routing and modular adaptation for multimodal continual instruction tuning.
Selected work
Publications & preprints
ProtoAda: Prototype-Guided Adaptive Adapter Expansion and Geometric Consolidation for Multimodal Continual Instruction Tuning
Preprint · arXiv:2606.02576
Recognition
Scholarships
2025
BYD Scholarship
2024