arXiv subjects

Heeju Kim

Publications and source records attributed to Heeju Kim.

6 recordsLinked to original sources

arXiv2026

Belief in Authority: Impact of Authority in Multi-Agent Evaluation Framework

Multi-agent systems utilizing large language models often assign authoritative roles to improve performance, yet the impact of authority bias on agent interactions remains underexplored. We present the first systematic analysis of role-based authority bias in free-form multi-agent evaluation using ChatEval. Applying French and Raven's power-based theory, we classify authoritative roles into legitimate, referent, and expert types and analyze their influence across 12-turn conversations. Experiments with GPT-4o and DeepSeek R1 reveal that Expert and Referent power roles exert stronger influence than Legitimate power roles. Crucially, authority bias emerges not through active conformity by general agents, but through authoritative roles consistently maintaining their positions while general agents demonstrate flexibility. Furthermore, authority influence requires clear position statements, as neutral responses fail to generate bias. These findings provide key insights for designing multi-agent frameworks with asymmetric interaction patterns.

cs.CL↗

arXiv2025

ReJump: A Tree-Jump Representation for Analyzing and Improving LLM Reasoning

Large Reasoning Models (LRMs) are Large Language Models (LLMs) explicitly trained to generate long-form Chain-of-Thoughts (CoTs), achieving impressive success on challenging tasks like math and programming. However, their underlying reasoning "algorithms" remain poorly understood. To investigate this, we propose ReJump, which represents a reasoning trace as a visitation order over nodes in a tree of intermediate problem-solving steps. Transitions between nodes, which we term jumps, include adjacent moves that capture behaviors such as calculation, and non-adjacent moves that capture behaviors such as backtracking and verification. ReJump enables analyzing LLM reasoning with diverse metrics that quantify exploration, exploitation, overthinking, forgetting, and verification. Using our proposed LLM agent to extract reasoning traces into ReJump format, we evaluate state-of-the-art LRMs on two tasks and find that models with similar accuracy can exhibit distinct reasoning behaviors, while different tasks favor different reasoning styles (e.g., varying balance between exploration and exploitation). To further understand how learning strategies shape reasoning, we use ReJump to compare distilled LRMs with their teachers, CoT-prompted LLMs with LRMs, and to examine how the number of reasoning examples and reinforcement learning affect reasoning behavior. Finally, we show that ReJump can improve reasoning quality at test time through strategies such as ReJump-guided Best-of-N selection and prompt selection. Our code is publicly available at https://github.com/UW-Madison-Lee-Lab/ReJump.

cs.LG↗

arXiv2025

Observation of Macroscopic Nonlocal Voltage at Room Temperature

Electrons in conductors suffer frequent scatterings with defects and phonons, and the diffusive macroscopic behaviors are determined by an external electric field through Ohms law. If electrons are no longer diffusive, the Ohmic description collapses. In devices composed of thin chalcogenides and YBa2Cu3O7, we observe a transition from an Ohmic conductor to a nonlocal conductor below a certain temperature. The nonlocal conductor is characterized by significant nonlocal voltages (~0.1 V) across macroscopic regions (~1 mm) that are conventionally considered to be equipotential. Nonlinear responses are an additional characteristic. Negative local resistances in a vicinal geometry support macroscopic hydrodynamic flow as the underlying mechanism, implying electron momentum conservation over incredibly long distances. This new conduction state, observable at room temperature, opens the field of nonlocal electronics and low-dissipation applications.

cond-mat.str-el↗

arXiv2025

VersaPRM: Multi-Domain Process Reward Model via Synthetic Reasoning Data

Process Reward Models (PRMs) have proven effective at enhancing mathematical reasoning for Large Language Models (LLMs) by leveraging increased inference-time computation. However, they are predominantly trained on mathematical data and their generalizability to non-mathematical domains has not been rigorously studied. In response, this work first shows that current PRMs have poor performance in other domains. To address this limitation, we introduce VersaPRM, a multi-domain PRM trained on synthetic reasoning data generated using our novel data generation and annotation method. VersaPRM achieves consistent performance gains across diverse domains. For instance, in the MMLU-Pro category of Law, VersaPRM via weighted majority voting, achieves a 7.9% performance gain over the majority voting baseline -- surpassing Qwen2.5-Math-PRM's gain of 1.3%. We further contribute to the community by open-sourcing all data, code and models for VersaPRM.

cs.LG↗

arXiv2025

Dynamic Modulation of Electronic and Optical Properties in GaN Bilayers by Interlayer Sliding

In this study, we present a first-principles investigation of the electronic and optical properties of gallium nitride (GaN) bilayers, focusing on the influence of interlayer sliding and spacing. In contrast to the earlier studies on discrete stacking configurations, we explore the dynamic evolution of the properties during transitions between stable stacking arrangements. Using density functional theory calculations, we systematically analyze the impact of these structural variations on the electronic band structure and optical absorption spectra of GaN bilayers. The analysis includes both high-symmetry stacking configurations (AA', AB', and AC') and intermediate states generated by controlled in-plane atomic displacements, thereby providing a comprehensive understanding of the property changes associated with interlayer sliding. The findings of this study provide valuable insights into the potential for tuning the electronic and optical response of two-dimensional GaN for applications in nanoscale photonic and electronic devices, where precise control over interlayer interactions and stacking is crucial.

cond-mat.mes-hall↗

arXiv2020

Adsorption properties of dopamine derivatives using carbon nanotubes: A first-principles study

Detecting dopamine is of great biological importance because the molecule plays many roles in the human body. For instance, the lack of dopamine release is the cause of Parkinson's disease. Although many researchers have carried out experiments on dopamine detection using carbon nanotubes (CNTs), there are only a few theoretical studies on this topic. We study the adsorption properties of dopamine and its derivatives, L-DOPA and dopamine o-quinone, adsorbed on a semiconducting (10, 0) CNT, using density functional theory calculations. Our computational simulations reveal that localized states originating from dopamine o-quinone appear in the bandgap of the (10, 0) CNT, but those originating from dopamine and L-DOPA do not appear in the gap. Therefore, dopamine o-quinone is expected to be detectable using an external electric field but dopamine and L-DOPA should be difficult to detect.

physics.app-ph↗