Searcharxiv⌕ Search

arXiv subjects

Jaehyeok Lee

Publications and source records attributed to Jaehyeok Lee.

14 recordsLinked to original sources
arXiv2026

When Noise Fabricates Bias: The Fragility of LLM-as-a-Judge Bias Measurement under Noisy Text

Large language models are increasingly used as judges to measure social bias in text, yet the passages they judge are often noisy, containing typos, informal spelling, and broken punctuation. The consequences of such surface noise for social bias measurement remain unclear. To investigate this question, we apply five realistic noise conditions at multiple intensity levels to 3,822 stereotype-related responses and compare the resulting bias judgments with those on the original text. We find that such surface noise does not degrade bias measurement symmetrically: it is far more likely to turn neutral judgments into biased ones than biased judgments into neutral ones, by up to a 120x margin. We further observe two non-obvious effects across four LLM judges: in the most fragile judge the distortion is at its purest at mild, realistic noise levels, where erasure is scarcest, and as judges grow robust it attenuates toward parity rather than reversing. Bias measured on noisy text is therefore systematically overestimated, most in the categories that matter most for fairness.

cs.CL↗
arXiv2026

Camellia: Benchmarking Cultural Biases in LLMs for Asian Languages

As Large Language Models (LLMs) develop stronger multilingual capabilities, their sensitivity to culturally diverse entities becomes increasingly important. Prior work by Naous et al. (2024) has shown that LLMs often favor Western-associated entities in Arabic. Due to the lack of entity-centric multilingual benchmarks, it remains unclear if such biases also manifest in various non-Western languages. In this paper, we introduce Camellia, a benchmark for evaluating entity-centric cultural biases in nine Asian languages, spanning six Asian cultures. Camellia includes 19,530 manually annotated entities associated with the covered Asian or Western cultures, as well as 2,173 masked contexts for these entities derived from social media posts. Using Camellia, we evaluate cultural biases in four recent multilingual LLMs across three tasks: cultural context adaptation, sentiment association, and entity extractive QA. Our analyses show that LLMs struggle with cultural adaptation across these languages, with performance differing across models developed in different regions. We further observe that different LLM families can hold distinct biases, reflected in the ways they link cultures to particular sentiments. Lastly, we find that LLMs can struggle with context understanding in some Asian languages, creating performance gaps between cultures in entity extraction.

cs.CL↗
arXiv2026

Topological signatures in the curvature-induced energy response

Relativistic effective field theory predicts a topological energy response to a gravitational field that appears at third order in spatial gradients. Here, we investigate how this response emerges in the nonrelativistic Haldane model using a microscopic lattice formulation of curvature-induced deformations. We find that the leading first-order energy response is nonuniversal and depends on the bond-resolved structure of the deformation; in particular, it vanishes for a symmetric modulation of the three nearest-neighbor hoppings. In contrast, the third-order response exhibits a discontinuity across the topological transition whose magnitude agrees with the relativistic gravitational Chern--Simons prediction. Thus, although the absolute response is nonuniversal, its third-order discontinuity is universal and retains a characteristic topological fingerprint beyond the relativistic limit.

cond-mat.mes-hall↗
arXiv2026

Distributional Open-Ended Evaluation of LLM Cultural Value Alignment Based on Value Codebook

As LLMs are globally deployed, aligning their cultural value orientations is critical for safety and user engagement. However, existing benchmarks face the Construct-Composition-Context ($C^3$) challenge: relying on discriminative, multiple-choice formats that probe value knowledge rather than true orientations, overlook subcultural heterogeneity, and mismatch with real-world open-ended generation. We introduce DOVE, a distributional evaluation framework that directly compares human-written text distributions with LLM-generated outputs. DOVE utilizes a rate-distortion variational optimization objective to construct a compact value codebook from 10K documents, mapping text into a structured value space to filter semantic noise. Alignment is measured using unbalanced optimal transport, capturing intra-cultural distributional structures and subgroup diversity. Experiments across 12 LLMs show that DOVE achieves superior predictive validity, attaining a 31.56% correlation with downstream tasks, while maintaining high reliability with as few as 500 samples per culture.

cs.CL↗
arXiv2026

SECOND-Grasp: Semantic Contact-guided Dexterous Grasping

Achieving reliable robotic manipulation, such as dexterous grasping, requires a synergy between physically stable interactions and semantic task guidance, yet these objectives are often treated as separate, disjoint goals. In this paper, we investigate how to integrate dexterous grasping techniques, i.e., physically stable grasps for object lifting and language-guided grasp generation, to achieve both physical stability and semantic understanding. To this end, we propose SECOND-Grasp (SEmantic CONtact-guided Dexterous Grasping), a unified framework that enables robotic hands to dynamically adjust grasping strategies based on semantic reasoning while ensuring physical feasibility. We begin by obtaining coarse contact proposals through vision-language reasoning to infer where contacts should occur based on object properties, followed by segmentation to localize these regions across views. To further ensure consistency across multiple viewpoints, we introduce Semantic-Geometric Consistency Refinement (SGCR), which refines initial contact predictions by enforcing semantic consistency across views and removing geometrically invalid regions, yielding reliable 3D contact maps. Then, we derive a feasible hand pose for each contact map via inverse kinematics, generating a supervision signal for policy learning. Our approach, trained on DexGraspNet, consistently outperforms baselines in lifting success rate on both seen and unseen categories, achieving 98.2% and 97.7%, respectively, while also improving intent-aware grasping by 12.8% and 26.2%. We further show promising results on additional datasets and robotic hands, including Shadow Hand and Allegro Hand.

cs.RO↗
arXiv2025

Tensor Enriched Categorical Generalization of the Eilenberg-Watts Theorem

Let $b$, $b'$ be commutative monoids in a Bénabou cosmos. Motivated by six-functor formalisms in algebraic geometry, we prove that the category of commutative monoids over $b\otimes b'$ is equivalent to the category of cocontinuous lax monoidal enriched functors between the monoidal enriched categories of right modules over $b$, $b'$.

math.CT↗
arXiv2025

Unintended Harms of Value-Aligned LLMs: Psychological and Empirical Insights

The application scope of Large Language Models (LLMs) continues to expand, leading to increasing interest in personalized LLMs that align with human values. However, aligning these models with individual values raises significant safety concerns, as certain values may correlate with harmful information. In this paper, we identify specific safety risks associated with value-aligned LLMs and investigate the psychological principles behind these challenges. Our findings reveal two key insights. (1) Value-aligned LLMs are more prone to harmful behavior compared to non-fine-tuned models and exhibit slightly higher risks in traditional safety evaluations than other fine-tuned models. (2) These safety issues arise because value-aligned LLMs genuinely generate text according to the aligned values, which can amplify harmful outcomes. Using a dataset with detailed safety categories, we find significant correlations between value alignment and safety risks, supported by psychological hypotheses. This study offers insights into the "black box" of value alignment and proposes in-context alignment methods to enhance the safety of value-aligned LLMs.

cs.CL↗
arXiv2025

Prekosmic Grothendieck/Galois Categories

We establish a generalized version of the duality between groups and the categories of their representations on sets. Given an abstract symmetric monoidal category $K$ called Galois prekosmos, we define pre-Galois objects in $K$ and study the categories of their representations internal to $K$. The motivating example of $K$ is the cartesian monoidal category $\textit{Set}$ of sets, and pre-Galois objects in $\textit{Set}$ are groups. We present an axiomatic definition of pre-Galois $K$-categories, which is a complete abstract characterization of the categories of representations of pre-Galois objects in $K$. The category of covering spaces over a well-connected topological space is a prototype of a pre-Galois $\textit{Set}$-category. We establish a perfect correspondence between pre-Galois objects in $K$ and pre-Galois $K$-categories pointed with pre-fiber functors. We also establish a generalized version of the duality between flat affine group schemes and the categories of their linear representations. Given an abstract symmetric monoidal category $K$ called Grothendieck prekosmos, we define what are pre-Grothendieck objects in $K$ and study the categories of their representations internal to $K$. The motivating example of $K$ is the symmetric monoidal category $\textit{Vec}_k$ of vector spaces over a field $k$, and pre-Grothendieck objects in $\textit{Vec}_k$ are affine group $k$-schemes. We present an axiomatic definition of pre-Grothendieck $K$-categories, which is a complete abstract characterization of the categories of representations of pre-Grothendieck objects in $K$. The indization of a neutral Tannakian category over a field $k$ is a prototype of a pre-Grothendieck $\textit{Vec}_k$-category. We establish a perfect correspondence between pre-Grothendieck objects in $K$ and pre-Grothendieck $K$-categories pointed with pre-fiber functors.

math.AG↗
arXiv2025

Self-Training Meets Consistency: Improving LLMs' Reasoning with Consistency-Driven Rationale Evaluation

Self-training approach for large language models (LLMs) improves reasoning abilities by training the models on their self-generated rationales. Previous approaches have labeled rationales that produce correct answers for a given question as appropriate for training. However, a single measure risks misjudging rationale quality, leading the models to learn flawed reasoning patterns. To address this issue, we propose CREST (Consistency-driven Rationale Evaluation for Self-Training), a self-training framework that further evaluates each rationale through follow-up questions and leverages this evaluation to guide its training. Specifically, we introduce two methods: (1) filtering out rationales that frequently result in incorrect answers on follow-up questions and (2) preference learning based on mixed preferences from rationale evaluation results of both original and follow-up questions. Experiments on three question-answering datasets using open LLMs show that CREST not only improves the logical robustness and correctness of rationales but also improves reasoning abilities compared to previous self-training approaches.

cs.LG↗
arXiv2022

Critical role of terminating layer in formation of 2DEG state at the $LaInO_{3}$/$BaSnO_{3}$ interface

Based on the interface polarization model, the two-dimensional electron gas (2DEG) at $LaInO_{3}$(LIO)/$BaSnO_{3}$(BSO) interfaces is understood to originate from a polarization discontinuity at the interface and the conduction band offset between LIO and BSO. In this scenario, the direction of polarization at the interface is determined by whether the first atomic LIO layer at the interface is LaO$^{+}$ or InO$_{2}^{-}$. We investigate the role of the terminating layer at the LIO/BSO interface in creating the 2DEG. Based on conductance measurements of our in-situ grown LIO/BSO heterostructures, we report in this work that the 2DEG only forms when the BSO surface is terminated with a SnO$_{2}$ layer. We controlled the terminating layer by additional SnO$_{2}$ deposition on the BSO surface. We show that the as-grown BSO surface has a mixed terminating layer of BaO and SnO$_{2}$ while the BSO surfaces prepared with additional SnO$_{2}$ deposition are terminated mainly with the SnO$_{2}$ layer. The terminating layer was confirmed by coaxial impact collision ion scattering spectroscopy (CAICISS). Our finding is consistent with the interface polarization model for 2DEG formation at LIO/BSO interfaces, in which the direction of the interfacial polarization in LIO is determined by the terminating layer of the BSO surface.

cond-mat.mtrl-sci↗
arXiv2021

Federated Learning: Issues in Medical Application

Since the federated learning, which makes AI learning possible without moving local data around, was introduced by google in 2017 it has been actively studied particularly in the field of medicine. In fact, the idea of machine learning in AI without collecting data from local clients is very attractive because data remain in local sites. However, federated learning techniques still have various open issues due to its own characteristics such as non identical distribution, client participation management, and vulnerable environments. In this presentation, the current issues to make federated learning flawlessly useful in the real world will be briefly overviewed. They are related to data/system heterogeneity, client management, traceability, and security. Also, we introduce the modularized federated learning framework, we currently develop, to experiment various techniques and protocols to find solutions for aforementioned issues. The framework will be open to public after development completes.

cs.LG↗
arXiv2019

Affine group dg-schemes and linear representations I - Basic theory and Tannakian reconstructions

We develop a basic theory of affine group dg-schemes, their Lie algebraic counterparts and linear representations. We prove Tannaka type reconstruction theorems that an affine group dg-scheme can be recovered from the dg-tensor category of its linear representations as well as from the rigid dg-tensor category of its finite dimensional linear representations along with the forgetful functors to the underlying dg-tensor category of cochain complexes.

math.AG↗
arXiv2019

Representable presheaves of groups on the homotopy category of cocommutative dg-coalgebras and Tannakian reconstruction

Motivated by rational homotopy theory, we study a representable presheaf of groups $\mathbf{\mathfrak{P}}$ on the homotopy category of cocommutative differential graded coalgebras, its Lie algebraic counterpart and its linear representations. We prove a Tannaka type reconstruction theorem that $\mathbf{\mathfrak{P}}$ can be recovered from the dg-category of its linear representations along with the forgetful dg-functor to the underlying dg-category of chain complexes.

math.AG↗