7 papers
Train for Many, Update with One: Improving Multi-Sample Reasoning via Fenchel-Bregman Posterior Learning
Sungjun Lim, Adrian Robert Minut, Nico Carlsen Daheim +3
Multi-sample reasoning measures candidate coverage, the probability that at least one of candidates is correct. However, minimizing the average loss of individual attempts does…
Reliable Self-Evolution with Imperfect Proxy Rewards
Kangjun Noh, Soyu Kim, Kyungwoo Song
Large language model (LLM)-based self-evolving search is a promising approach to scientific discovery. However, high-fidelity evaluation of every candidate is prohibitively expensi…
Geometry-Adaptive Explainer for Faithful Dictionary-Based Interpretability under Distribution Shift
Sungjun Lim, Heedong Kim, Andrew Lee +1
Mechanistic interpretability aims to explain a model's behavior by identifying causally responsible internal structures. Dictionary-based explainers such as sparse autoencoders and…
Multi-LLM Adaptive Conformal Inference for Reliable LLM Responses
Kangjun Noh, Seongchan Lee, Ilmun Kim +1
Ensuring factuality is essential for the safe use of Large Language Models (LLMs) in high-stakes domains such as medicine and law. Conformal inference provides distribution-free gu…
Semi-Supervised Preference Optimization with Limited Feedback
Seonggyun Lee, Sungjun Lim, Seojin Park +2
The field of preference optimization has made outstanding contributions to the alignment of language models with human preferences. Despite these advancements, recent methods still…
Eigen-Value: Efficient Domain-Robust Data Valuation via Eigenvalue-Based Approach
Youngjun Choi, Joonseong Kang, Sungjun Lim +1
Data valuation has become central in the era of data-centric AI. It drives efficient training pipelines and enables objective pricing in data markets by assigning a numeric value t…
Uncertainty-driven Embedding Convolution
Sungjun Lim, Kangjun Noh, Youngjun Choi +2
Text embeddings are essential components in modern NLP pipelines. Although numerous embedding models have been proposed, no single model consistently dominates across domains and t…