Publications | 南京大学大模型研究协同创新中心 https://cs.nju.edu.cn/lm/en/publication/ Publications Hugo Blox Builder (https://hugoblox.com)enFri, 08 May 2026 00:00:00 +0000 https://cs.nju.edu.cn/lm/media/icon_hu_6873430e16214d30.png Publications https://cs.nju.edu.cn/lm/en/publication/ A Data-Efficient Path to Multilingual LLMs: Language Expansion via Post-training PARAM$Δ$ Integration into Upcycled MoE https://cs.nju.edu.cn/lm/en/publication/zhou-data-efficient-2026/ Fri, 08 May 2026 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/zhou-data-efficient-2026/ AEA: Adaptive Expert Allocation Improves Sentence Embeddings from Mixture-of-Experts LLM https://cs.nju.edu.cn/lm/en/publication/yang-aea-2026/ Fri, 08 May 2026 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/yang-aea-2026/ Bootstrapping Code Translation with Weighted Multilanguage Exploration https://cs.nju.edu.cn/lm/en/publication/wu-bootstrapping-2026/ Fri, 08 May 2026 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/wu-bootstrapping-2026/ Focusing Condition: Inference-Time Self-Contrastive Steering Elicits Better Conditional Text Embeddings in LLMs https://cs.nju.edu.cn/lm/en/publication/cheng-focusing-2026/ Fri, 08 May 2026 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/cheng-focusing-2026/ How Do Answer Tokens Read Reasoning Traces? Self-Reading Patterns in Thinking LLMs for Quantitative Reasoning https://cs.nju.edu.cn/lm/en/publication/chen-answer-2026/ Fri, 08 May 2026 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/chen-answer-2026/ PEGRL: Improving Machine Translation by Post-Editing Guided Reinforcement Learning https://cs.nju.edu.cn/lm/en/publication/shen-pegrl-2026/ Fri, 08 May 2026 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/shen-pegrl-2026/ Reasoning While Asking: Transforming Reasoning Large Language Models from Passive Solvers to Proactive Inquirers https://cs.nju.edu.cn/lm/en/publication/chen-reasoning-2026/ Fri, 08 May 2026 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/chen-reasoning-2026/ To Diff or Not to Diff? Structure-Aware and Adaptive Output Formats for Efficient LLM-based Code Editing https://cs.nju.edu.cn/lm/en/publication/cheng-diff-2026/ Fri, 08 May 2026 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/cheng-diff-2026/ Understanding New-Knowledge-Induced Factual Hallucinations in LLMs: Analysis and Interpretation https://cs.nju.edu.cn/lm/en/publication/dang-understanding-2026/ Fri, 08 May 2026 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/dang-understanding-2026/ AV-Reasoner: Improving and Benchmarking Clue-Grounded Audio-Visual Counting for MLLMs https://cs.nju.edu.cn/lm/en/publication/lu-av-reasoner-2026-cvpr/ Tue, 05 May 2026 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/lu-av-reasoner-2026-cvpr/ Bayesian Decomposition and Semantic Completion for Few-shot Semantic Segmentation https://cs.nju.edu.cn/lm/en/publication/shi-bayesian-2026-cvpr/ Tue, 05 May 2026 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/shi-bayesian-2026-cvpr/ CoMo: Learning Continuous Latent Motion from Internet Videos for Scalable Robot Learning https://cs.nju.edu.cn/lm/en/publication/yang-co-mo-2026-cvpr/ Tue, 05 May 2026 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/yang-co-mo-2026-cvpr/ DDT: Decoupled Diffusion Transformer https://cs.nju.edu.cn/lm/en/publication/wang-ddt-2026-cvpr/ Tue, 05 May 2026 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/wang-ddt-2026-cvpr/ InternVideo-Next: Towards World-Understanding Video Models https://cs.nju.edu.cn/lm/en/publication/wang-intern-video-next-2026-cvpr/ Tue, 05 May 2026 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/wang-intern-video-next-2026-cvpr/ Rethinking BCE Loss for Multi-Label Image Recognition with Fine-Tuning https://cs.nju.edu.cn/lm/en/publication/zhou-rethinking-bce-2026-cvpr/ Tue, 05 May 2026 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/zhou-rethinking-bce-2026-cvpr/ TempR1: Improving Temporal Understanding of MLLMs via Temporal-Aware Multi-Task Reinforcement Learning https://cs.nju.edu.cn/lm/en/publication/wu-temp-r-1-2026-cvpr/ Tue, 05 May 2026 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/wu-temp-r-1-2026-cvpr/ TimeLens: Rethinking Video Temporal Grounding with Multimodal LLMs https://cs.nju.edu.cn/lm/en/publication/zhang-time-lens-2026-cvpr/ Tue, 05 May 2026 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/zhang-time-lens-2026-cvpr/ UniAVGen: Unified Audio and Video Generation with Asymmetric Cross-Modal Interactions https://cs.nju.edu.cn/lm/en/publication/zhang-uni-av-gen-2026-cvpr/ Tue, 05 May 2026 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/zhang-uni-av-gen-2026-cvpr/ VideoRealBench: A Chain-of-Thought Realism Evaluation Benchmark for Generated Human-Centric Videos https://cs.nju.edu.cn/lm/en/publication/yang-video-real-bench-2026-cvpr/ Tue, 05 May 2026 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/yang-video-real-bench-2026-cvpr/ VMonarch: Efficient Video Diffusion Transformers with Structured Attention https://cs.nju.edu.cn/lm/en/publication/liang-v-monarch-2026-cvpr/ Tue, 05 May 2026 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/liang-v-monarch-2026-cvpr/ Will Multimodal Models Be Dazzled by Multi-Image Visual Puzzles? https://cs.nju.edu.cn/lm/en/publication/zhu-multimodal-puzzles-2026-cvpr/ Tue, 05 May 2026 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/zhu-multimodal-puzzles-2026-cvpr/ Arbitrary Generative Video Interpolation https://cs.nju.edu.cn/lm/en/publication/zhang-arbitrary-2026/ Fri, 24 Apr 2026 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/zhang-arbitrary-2026/ Balancing the Experts: Unlocking LoRA-MoE for GRPO via Mechanism-Aware Rewards https://cs.nju.edu.cn/lm/en/publication/ma-balancing-2026/ Fri, 24 Apr 2026 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/ma-balancing-2026/ CaReBench: A Fine-grained Benchmark for Video Captioning and Retrieval https://cs.nju.edu.cn/lm/en/publication/xu-carebench-2026/ Fri, 24 Apr 2026 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/xu-carebench-2026/ DuPO: Enabling Reliable Self-Verification via Dual Preference Optimization https://cs.nju.edu.cn/lm/en/publication/she-dupo-2026/ Fri, 24 Apr 2026 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/she-dupo-2026/ Long-Context Attention Benchmark: From Kernel Efficiency to Distributed Context Parallelism https://cs.nju.edu.cn/lm/en/publication/bu-long-context-2026/ Fri, 24 Apr 2026 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/bu-long-context-2026/ PixNerd: Pixel Neural Field Diffusion https://cs.nju.edu.cn/lm/en/publication/wang-pixnerd-2026/ Fri, 24 Apr 2026 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/wang-pixnerd-2026/ PoseX: AI Defeats Physics-based Methods on Protein Ligand Cross-Docking https://cs.nju.edu.cn/lm/en/publication/jiang-posex-2026/ Fri, 24 Apr 2026 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/jiang-posex-2026/ RIVER: Real-time Video Interaction Benchmark https://cs.nju.edu.cn/lm/en/publication/shi-river-2026/ Fri, 24 Apr 2026 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/shi-river-2026/ The First Impression Problem: Internal Bias Triggers Overthinking in Reasoning Models https://cs.nju.edu.cn/lm/en/publication/dang-first-2026/ Fri, 24 Apr 2026 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/dang-first-2026/ UniFlow: A Unified Pixel Flow Tokenizer for Visual Understanding and Generation https://cs.nju.edu.cn/lm/en/publication/yue-uniflow-2026/ Fri, 24 Apr 2026 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/yue-uniflow-2026/ VideoChat-Flash: Hierarchical Compression for Long-Context Video Modeling https://cs.nju.edu.cn/lm/en/publication/li-videochat-flash-2026/ Fri, 24 Apr 2026 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/li-videochat-flash-2026/ 3D interaction geometric pre-training for molecular relational learning https://cs.nju.edu.cn/lm/en/publication/lee-3-d-2025/ Sat, 11 Oct 2025 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/lee-3-d-2025/ Eagle 2.5: boosting long-context post-training for frontier vision-language models https://cs.nju.edu.cn/lm/en/publication/chen-eagle-2025/ Sat, 11 Oct 2025 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/chen-eagle-2025/ EgoExoBench: a benchmark for first-and third-person view video understanding in MLLMs https://cs.nju.edu.cn/lm/en/publication/he-egoexobench-2025/ Sat, 11 Oct 2025 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/he-egoexobench-2025/ Gated integration of low-rank adaptation for continual learning of language models https://cs.nju.edu.cn/lm/en/publication/liang-gated-2025/ Sat, 11 Oct 2025 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/liang-gated-2025/ LongVPO: from anchored cues to self-reasoning for long-form video preference optimization https://cs.nju.edu.cn/lm/en/publication/huang-longvpo-2025/ Sat, 11 Oct 2025 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/huang-longvpo-2025/ Loquetier: a virtualized multi-LoRA framework for unified LLM fine-tuning and serving https://cs.nju.edu.cn/lm/en/publication/zhang-loquetier-2025/ Sat, 11 Oct 2025 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/zhang-loquetier-2025/ MotionRAG: motion retrieval-augmented image-to-video generation https://cs.nju.edu.cn/lm/en/publication/zhu-motionrag-2025/ Sat, 11 Oct 2025 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/zhu-motionrag-2025/ StreamForest: efficient online video understanding with persistent event memory https://cs.nju.edu.cn/lm/en/publication/zeng-streamforest-2025/ Sat, 11 Oct 2025 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/zeng-streamforest-2025/ VideoChat-r1.5: visual test-time scaling to reinforce multimodal reasoning by iterative perception https://cs.nju.edu.cn/lm/en/publication/yan-videochat-2025/ Sat, 11 Oct 2025 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/yan-videochat-2025/ Correspondence as video: test-time adaption on SAM2 for reference segmentation in the wild https://cs.nju.edu.cn/lm/en/publication/wang-correspondence-2025/ Tue, 12 Aug 2025 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/wang-correspondence-2025/ Divide-and-conquer for enhancing unlabeled learning, stability, and plasticity in semi-supervised continual learning https://cs.nju.edu.cn/lm/en/publication/duan-divide-2025/ Tue, 12 Aug 2025 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/duan-divide-2025/ Make your training flexible: towards deployment-efficient video models https://cs.nju.edu.cn/lm/en/publication/wang-make-2025/ Tue, 12 Aug 2025 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/wang-make-2025/ MobileViCLIP: an efficient video-text model for mobile devices https://cs.nju.edu.cn/lm/en/publication/yang-mobileviclip-2025/ Tue, 12 Aug 2025 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/yang-mobileviclip-2025/ p-MoD: building mixture-of-depths MLLMs via progressive ratio decay https://cs.nju.edu.cn/lm/en/publication/zhang-p-mod-2025/ Tue, 12 Aug 2025 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/zhang-p-mod-2025/ Scalable image tokenization with index backpropagation quantization https://cs.nju.edu.cn/lm/en/publication/shi-scalable-2025/ Tue, 12 Aug 2025 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/shi-scalable-2025/ VRBench: a benchmark for multi-step reasoning in long narrative videos https://cs.nju.edu.cn/lm/en/publication/yu-vrbench-2025/ Tue, 12 Aug 2025 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/yu-vrbench-2025/ Differentiable solver search for fast diffusion sampling https://cs.nju.edu.cn/lm/en/publication/wang-differentiable-2025/ Fri, 18 Jul 2025 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/wang-differentiable-2025/ Elucidating the design space of multimodal protein language models https://cs.nju.edu.cn/lm/en/publication/wang-elucidating-2025/ Fri, 18 Jul 2025 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/wang-elucidating-2025/ On the tension between Byzantine robustness and no-attack accuracy in distributed learning https://cs.nju.edu.cn/lm/en/publication/yang-tension-2025/ Fri, 18 Jul 2025 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/yang-tension-2025/ Stochastic layer-wise shuffle for improving Vision Mamba training https://cs.nju.edu.cn/lm/en/publication/huang-stochastic-2025/ Fri, 18 Jul 2025 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/huang-stochastic-2025/ AutoLUT: LUT-based image super-resolution with automatic sampling and adaptive residual learning https://cs.nju.edu.cn/lm/en/publication/xu-autolut-2025/ Sun, 20 Apr 2025 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/xu-autolut-2025/ Balanced direction from multifarious choices: arithmetic meta-learning for domain generalization https://cs.nju.edu.cn/lm/en/publication/wang-balanced-2025/ Sun, 20 Apr 2025 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/wang-balanced-2025/ CATANet: efficient content-aware token aggregation for lightweight image super-resolution https://cs.nju.edu.cn/lm/en/publication/liu-catanet-2025/ Sun, 20 Apr 2025 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/liu-catanet-2025/ Contextual AD narration with interleaved multimodal sequence https://cs.nju.edu.cn/lm/en/publication/wang-contextual-2025/ Sun, 20 Apr 2025 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/wang-contextual-2025/ LeviTor: 3D trajectory oriented image-to-video synthesis https://cs.nju.edu.cn/lm/en/publication/wang-levitor-2025/ Sun, 20 Apr 2025 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/wang-levitor-2025/ Multiple object tracking as id prediction https://cs.nju.edu.cn/lm/en/publication/gao-multiple-2025/ Sun, 20 Apr 2025 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/gao-multiple-2025/ Online video understanding: a comprehensive benchmark and memory-augmented method https://cs.nju.edu.cn/lm/en/publication/huang-online-2025/ Sun, 20 Apr 2025 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/huang-online-2025/ Steady progress beats stagnation: mutual aid of foundation and conventional models in mixed domain semi-supervised medical image segmentation https://cs.nju.edu.cn/lm/en/publication/ma-steady-2025/ Sun, 20 Apr 2025 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/ma-steady-2025/ Task preference optimization: improving multimodal large language models with vision task alignment https://cs.nju.edu.cn/lm/en/publication/yan-task-2025/ Sun, 20 Apr 2025 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/yan-task-2025/ Taste more, taste better: diverse data and strong model boost semi-supervised crowd counting https://cs.nju.edu.cn/lm/en/publication/yang-taste-2025/ Sun, 20 Apr 2025 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/yang-taste-2025/ Tra-MoE: learning trajectory prediction model from multiple domains for adaptive policy conditioning https://cs.nju.edu.cn/lm/en/publication/yang-tra-moe-2025/ Sun, 20 Apr 2025 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/yang-tra-moe-2025/ Uniap: unifying inter-and intra-layer automatic parallelism by mixed integer quadratic programming https://cs.nju.edu.cn/lm/en/publication/lin-uniap-2025/ Sun, 20 Apr 2025 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/lin-uniap-2025/ CG-Bench: Clue-grounded Question Answering Benchmark for Long Video Understanding https://cs.nju.edu.cn/lm/en/publication/chen-cg-bench-2025/ Tue, 15 Apr 2025 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/chen-cg-bench-2025/ MeteoRA: Multiple-tasks Embedded LoRA for Large Language Models https://cs.nju.edu.cn/lm/en/publication/xu-meteora-2025/ Tue, 15 Apr 2025 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/xu-meteora-2025/ Proving Olympiad Inequalities by Synergizing LLMs and Symbolic Reasoning https://cs.nju.edu.cn/lm/en/publication/li-proving-2025/ Tue, 15 Apr 2025 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/li-proving-2025/ SPA: 3D Spatial-Awareness Enables Effective Embodied Representation https://cs.nju.edu.cn/lm/en/publication/zhu-spa-2025/ Tue, 15 Apr 2025 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/zhu-spa-2025/ TimeSuite: Improving MLLMs for Long Video Understanding via Grounded Tuning https://cs.nju.edu.cn/lm/en/publication/zeng-timesuite-2025/ Tue, 15 Apr 2025 00:00:00 +0000 https://cs.nju.edu.cn/lm/en/publication/zeng-timesuite-2025/