10 papers
Answer First, Reason Later: When Commitment Order Costs Accuracy in Diffusion Language Models
Jewon Yeom, Jaewon Sok, Seonghyeon Park +3
Masked diffusion language models revise many masked output positions in parallel. We call a token committed once it becomes visible and is never masked again, and call a response a…
Inference-Time Vulnerability Beyond Shallow Safety: Alignment Along Generation Trajectories
Kyungmin Park, Taesup Kim
Safety-aligned Large Language Models (LLMs) remain vulnerable to interventions during inference that redirect generation toward harmful outputs. Recent work attributes this to shal…
Hallucination as Commitment Failure: Larger LLMs Misfire Despite Knowing the Answer
Jewon Yeom, Jaewon Sok, Heejun Kim +3
Hallucination is often viewed as a direct consequence of missing knowledge: a model answers incorrectly when the correct answer is absent from its generation-time distribution, and…
From Noise to Diversity: Random Embedding Injection in LLM Reasoning
Heejun Kim, Seungpil Lee, Jewon Yeom +5
Recent soft prompt research has tried to improve reasoning by inserting trained vectors into LLM inputs, yet whether the gain comes from the learned content or from the act of inje…
Two Birds, One Projection: Harmonizing Safety and Utility in LVLMs via Inference-time Feature Projection
Yewon Han, Yumin Seol, EunGyung Kong +2
Existing jailbreak defence frameworks for Large Vision-Language Models often suffer from a safety utility tradeoff, where strengthening safety inadvertently degrades performance on…
Efficient Epistemic Uncertainty Estimation for Large Language Models via Knowledge Distillation
Seonghyeon Park, Jewon Yeom, Jaewon Sok +3
Quantifying uncertainty in Large Language Models (LLMs) is essential for mitigating hallucinations and enabling risk-aware deployment in safety-critical tasks. However, estimating…
Garbage Attention in Large Language Models: BOS Sink Heads and Sink-aware Pruning
Jaewon Sok, Jewon Yeom, Seonghyeon Park +2
Large Language Models (LLMs) are known to contain significant redundancy, yet a systematic explanation for why certain components, particularly in higher layers, are more redundant…
EpiCaR: Knowing What You Don't Know Matters for Better Reasoning in LLMs
Jewon Yeom, Jaewon Sok, Seonghyeon Park +2
Improving the reasoning abilities of large language models (LLMs) has largely relied on iterative self-training with model-generated data. While effective at boosting accuracy, exi…
Stable On-Policy Distillation through Adaptive Target Reformulation
Ijun Jang, Jewon Yeom, Juan Yeo +2
Knowledge distillation (KD) is a widely adopted technique for transferring knowledge from large language models to smaller student models; however, conventional supervised KD often…
Generalized and Personalized Federated Learning with Black-Box Foundation Models via Orthogonal Transformations
Eun Gyung Kong, Je Won Yeom, Yonghoon Jeon +1
Federated Learning (FL) facilitates decentralized model training while preserving data privacy. However, achieving both robust generalization and effective personalization simultan…