16 papers
Condition-Contrastive Extrapolation for On-Policy Distillation of Language Models
Wenxuan Luo, Chenxi Li, Haoyue Liu +2
On-policy distillation (OPD) improves a student language model by training on its own generated trajectories with supervision from a stronger teacher. However, when conditioned on…
How Should a Prompt Optimizer Spend a Tight Budget? BudgetAPO with Noise-Adaptive Evaluation
Haoyue Liu, Zhichao Wang, Huanyu Yan +1
Automatic prompt optimization (APO) has been widely employed to adapt large language models without updating their weights, yielding promising results. However, existing methods su…
How Should Teachers Be Prepared? RL on Student-Induced States for On-Policy Distillation
Xiaoyu Ma, Haoyue Liu, Zhichao Wang +2
On-policy distillation (OPD) improves the reasoning capabilities of small language models through token-level teacher supervision on student-generated trajectories. Yet can teacher…
Do Language Models Need Music Supervision? Verifiable Rewards for Multi-Constraint Symbolic Music Generation
Haoyue Liu, Ye Chen, Zhichao Wang +3
Language models can now generate symbolic music from natural-language instructions. However, existing models still struggle to satisfy multiple explicit musical constraints simulta…
Why Sample What You Can Enumerate? Exact Policy Optimization for Genomic Tool Selection
Haoyue Liu, Xiaoyu Ma, Ye Chen +2
Reinforcement learning over a frozen reasoner has become a common recipe for teaching a policy which external tools to invoke. We show that this recipe becomes structurally mismatc…
SEPO: Evidence-Grounded Prompt Optimization via Structural Editing
Xiaoyu Ma, Haoyue Liu, Yiwen Li +4
Existing API-only prompt optimisers are often described as interpretable, but in practice, this usually means only post-hoc inspectability: each iteration still rewrites the prompt…
Do SpeechLMs Hear Their Own Opinions? Diagnosing and Mitigating Previous-Belief Contamination in Streaming Emotion Understanding
Haoyue Liu, Zhichao Wang, Ye Chen +2
Streaming emotion understanding uses historical state while continuously interpreting current audio, often feeding the model's previous prediction back as context. We show that thi…
Which Negatives Matter? Ask Your Text Encoder: Adaptive Similarity Margins for Dense-Caption Retrieval
Haoyue Liu, Ye Chen, Zhichao Wang +1
Dense-caption retrieval has recently been improved by introducing segmentation, edge maps, LLM-filtered captions, and cross-modal modules into contrastive fine-tuning. However, the…
Are Prompt Optimizers Blind? Cross-Modal Visual Feedback for Automatic Prompt Optimization
Haoyue Liu, Xiaoyu Ma, Ye Chen +2
Automatic prompt optimization (APO) has been widely adopted to adapt vision-language models (VLMs) to downstream tasks without weight updates, yielding promising results. However,…
One Rewrite to Fix Them All? Type-Aware Repair Allocation for Text-to-Image Prompt Optimization
Haoyue Liu, Xiaoyu Ma, Ye Chen +2
Text-to-image (T2I) generators often fail to follow their prompts faithfully, producing wrong counts, swapped attributes, ambiguous relations, and illegible text. Prompt optimizati…
SIGMA: Skill-Incidence Graphs for Compositional Multi-Agent Design
Kun Zeng, Yu Huo, Siyu Zhang +5
Existing graph-based multi-agent system (MAS) designers mainly improve collaboration by optimizing communication topologies over predefined agents, roles, or groups. However, becau…
Voices Across Registers: Corpus-Conditioned Vernacular Jailbreaks against Aligned LLMs via Fanfiction Subgenres
Zhongze Luo, Ruihe Shi, Zhenshuai Yin +3
Existing jailbreaks against aligned LLMs are discrete artifacts whose surface forms are easy to fingerprint and patch. We argue that the broader failure mode may lie not in any spe…
Select Smarter, Not More: Prompt-Aware Evaluation Scheduling with Submodular Guarantees
Xiaoyu Ma, Yiwen Li, Haoyue Liu +4
Automatic prompt optimization (APO) hinges on the quality of its evaluation signal, yet scoring every prompt candidate on the full training set is prohibitively expensive. Existing…
Adaptive Prompt Structure Factorization: A Framework for Self-Discovering and Optimizing Compositional Prompt Programs
Haoyue Liu, Zhichao Wang, Yongxin Guo +2
Automated prompt optimization is crucial for eliciting reliable reasoning from large language models (LLMs), yet most API-only prompt optimizers iteratively edit monolithic prompts…
PIDP-Attack: Combining Prompt Injection with Database Poisoning Attacks on Retrieval-Augmented Generation Systems
Haozhen Wang, Haoyue Liu, Jionghao Zhu +3
Large Language Models (LLMs) have demonstrated remarkable performance across a wide range of applications. However, their practical deployment is often hindered by issues such as o…
Shape of Thought: Progressive Object Assembly via Visual Chain-of-Thought
Yu Huo, Siyu Zhang, Kun Zeng +7
Multimodal models for text-to-image generation have achieved strong visual fidelity, yet they remain brittle under compositional structural constraints, notably generative numeracy…