arXiv subjects

Soyoung Oh

Publications and source records attributed to Soyoung Oh.

10 recordsLinked to original sources

arXiv2026

A retrieval conditioned rebinding circuit for dynamic entity tracking in large language models

To interpret context correctly and retrieve relevant information, large language models must bind entities to their attributes and update these bindings as state changes. We analyze how LLMs implement this binding process in a dynamic state tracking. Using causal interventions, we identify a retrieval conditioned rebinding mechanism, a compact attention head circuit that propagated binding information and when the entity is queried, uses the updated binding to retrieve the corresponding attribute. Across Gemma and Llama models, this circuit supports rebinding behavior, but the representational signature of the mechanism differs across model families. In Gemma models, the binding signature is clearly expressed in the query/key subspaces of the relevant attention heads, whereas in Llama models, the binding information is carried primarily in key vectors. Overall, our results reveal an interpretable mechanism for context dependent state tracking in LLMs.

cs.CL↗

arXiv2026

Beyond Routing Weights: Faithful Response-Level Interpretation of Mixture-of-Experts Reward Models via Contribution Contrast

Reward models are central to learning from human preferences, yet identifying what drives their predictions remains challenging. Recent sparse Mixture-of-Experts (MoE) reward models seek to improve interpretability by routing prompts to specialized experts and characterizing experts through examples with high routing weights. However, routing weights only reveal which prompts an expert $\textit{receives}$, not how it $\textit{judges}$ responses, providing only a partial account of expert behavior. We therefore propose $\textbf{Co}$ntribution-$\textbf{Co}$ntrast ($\textbf{CoCo}$) response-level interpretation, which faithfully characterizes experts' roles using chosen-rejected response pairs with the largest contribution contrasts, jointly capturing routing and preference behavior. Across automatic and human evaluations, CoCo yields more coherent, faithful, and specialized interpretations than router-based, score-based, and sparse autoencoder-based alternatives while maintaining competitive reward modeling accuracy. To the best of our knowledge, this is the first systematic study of interpretation methods for MoE reward models.

cs.AI↗

arXiv2026

Sparse Mixture-of-Experts Reward Models Learn Interpretable and Specialized Experts for Personalized Preference Modeling

Preference modeling plays a central role in reinforcement learning from human feedback (RLHF), enabling large language models (LLMs) to align with human values. However, most existing approaches assume a universal reward function, neglecting the diversity and heterogeneity of human preferences. To address this limitation without additional annotation costs, recent work has proposed learning multiple preference components from binary data and combining them to model individual preferences. Nevertheless, these components often fail to capture coherent and disentangled patterns, limiting their interpretability and effectiveness for personalization. In this work, we propose a sparse Mixture-of-Experts (MoE) reward model that encourages sparse routing and expert diversity during training on binary preference data. Across controlled and real-world experiments, sparse MoE learns interpretable routing patterns and specialized experts. It also improves test-time personalization, and post-adaptation shifts in expert weights provide a qualitative lens for analyzing how the model adapts to personalized preferences.

cs.LG↗

arXiv2026

Bridging Fairness and Explainability: Can Input-Based Explanations Promote Fairness in Hate Speech Detection?

Natural language processing (NLP) models often replicate or amplify social bias from training data, raising concerns about fairness. At the same time, their black-box nature makes it difficult for users to recognize biased predictions and for developers to effectively mitigate them. While some studies suggest that input-based explanations can help detect and mitigate bias, others question their reliability in ensuring fairness. Existing research on explainability in fair NLP has been predominantly qualitative, with limited large-scale quantitative analysis. In this work, we conduct the first systematic study of the relationship between explainability and fairness in hate speech detection, focusing on both encoder- and decoder-only models. We examine three key dimensions: (1) identifying biased predictions, (2) selecting fair models, and (3) mitigating bias during model training. Our findings show that input-based explanations can effectively detect biased predictions and serve as useful supervision for reducing bias during training, but they are unreliable for selecting fair models among candidates.Our code is available at https://github.com/Ewanwong/fairness_x_explainability.

cs.CL↗

arXiv2026

Tug-of-war between idioms' figurative and literal interpretations in LLMs

Idioms present a unique challenge for language models due to their non-compositional figurative interpretations, which often strongly diverge from the idiom's literal interpretation. In this paper, we employ causal tracing to systematically analyze how pretrained causal transformers deal with this ambiguity. We localize three mechanisms: (i) Early sublayers and specific attention heads retrieve an idiom's figurative interpretation, while suppressing its literal interpretation. (ii) When disambiguating context precedes the idiom, the model leverages it from the earliest layer and later layers refine the interpretation if the context conflicts with the retrieved interpretation. (iii) Then, selective, competing pathways carry both interpretations: an intermediate pathway prioritizes the figurative interpretation and a parallel direct route favors the literal interpretation, ensuring that both readings remain available. Our findings provide mechanistic evidence for idiom comprehension in autoregressive transformers.

cs.CL↗

arXiv2025

Tailoring the Electronic Configurations of YPc$_2$ on Cu(111): Decoupling Strategies for Molecular Spin Qubits Platforms

Among the potential spin qubit candidates, yttrium phthalocyanine double-decker (YPc$_2$) features a diamagnetic metal ion core that stabilizes the molecular structure, while its magnetic properties arise primarily from an unpaired electron (S=1/2) delocalized over the two phthalocyanine (Pc) ligands. Understanding its properties in the proximity of metal electrodes is crucial to assess its potential use in molecular spin qubits architectures. Here, we investigated the morphology and electronic structure of this molecule adsorbed on Cu(111) surface using scanning tunneling microscopy (STM). On Cu(111), YPc$_2$ adsorbs flat, with isolated molecules showing a preferred orientation along the <111> crystal axes. Moreover, we observed two different types of self-assembly molecular packing when growing molecular patches. For YPc$_2$ in direct contact with Cu(111), STM revealed a widely separated highest occupied and lowest unoccupied molecular orbitals (HOMO/LUMO), suggesting the quenching of the unpaired spin. Conversely, when YPc$_2$ is separated from the metal substrate by a few-layer thick diamagnetic zinc phthalocyanine (ZnPc) layer, we found the HOMO to split into singly occupied and singly unoccupied molecular orbitals (SOMO/SUMO). We observed that more than 2 layers of ZnPc are needed to avoid intermixing between the two molecules and spin quenching in YPc2. Density functional theory (DFT) reveals the spin quenching is due to the hybridization between YPc$_2$ and Cu(111) states, confirming the importance of using suitable decoupling layers to preserve the unpaired molecular spin. Our results suggest the potential of YPc$_2$/ZnPc heterostructures as a stable and effective molecular spin qubit platform and validate the possibility of integrating this molecular spin qubit candidate in future quantum logic devices.

cond-mat.mes-hall↗

arXiv2024

A Survey on the Role of Crowds in Combating Online Misinformation: Annotators, Evaluators, and Creators

Online misinformation poses a global risk with significant real-world consequences. To combat misinformation, current research relies on professionals like journalists and fact-checkers for annotating and debunking misinformation, and develops automated machine learning methods for detecting misinformation. Complementary to these approaches, recent research has increasingly concentrated on utilizing the power of ordinary social media users, a.k.a. "crowd", who act as eyes-on-the-ground proactively questioning and countering misinformation. Notably, recent studies show that 96% of counter-misinformation responses originate from them. Acknowledging their prominent role, we present the first systematic and comprehensive survey of research papers that actively leverage the crowds to combat misinformation. We first identify 88 papers related to crowd-based efforts, following a meticulous annotation process adhering to the PRISMA framework. We then present key statistics related to misinformation, counter-misinformation, and crowd input in different formats and topics. Upon holistic analysis of the papers, we introduce a novel taxonomy of the roles played by the crowds: (i)annotators who actively identify misinformation; (ii)evaluators who assess counter-misinformation effectiveness; (iii)creators who create counter-misinformation. This taxonomy explores the crowd's capabilities in misinformation detection, identifies prerequisites for effective counter-misinformation, and analyzes crowd-generated counter-misinformation. Then, we delve into (i)distinguishing individual, collaborative, and machine-assisted labeling for annotators; (ii)analyzing the effectiveness of counter-misinformation through surveys, interviews, and in-lab experiments for evaluators; and (iii)characterizing creation patterns and creator profiles for creators. Finally, we outline potential future research in this field.

cs.SI↗

arXiv2024

A continuous-wave and pulsed X-band electron spin resonance spectrometer operating in ultra-high vacuum for the study of low dimensional spin ensembles

We report the development of a continuous-wave and pulsed X-band electron spin resonance (ESR) spectrometer for the study of spins on ordered surfaces down to cryogenic temperatures. The spectrometer operates in ultra-high vacuum and utilizes a half-wavelength microstrip line resonator realized using epitaxially grown copper films on single crystal Al$_2$O$_3$ substrates. The one-dimensional microstrip line resonator exhibits a quality factor of more than 200 at room temperature, close to the upper limit determined by radiation losses. The surface characterizations of the copper strip of the resonator by atomic force microscope, low-energy electron diffraction, and scanning tunneling microscope show that the surface is atomically clean, flat, and single crystalline. Measuring the ESR spectrum at 15 K from a few nm thick molecular film of YPc$_2$, we find a continuous-wave ESR sensitivity of $2.6 \cdot 10^{11}~\text{spins}/\text{G} \cdot \text{Hz}^{1/2}$ indicating that a signal-to-noise ratio of $3.9~\text{G} \cdot \text{Hz}^{1/2}$ is expected from a monolayer of YPc$_2$ molecules. Advanced pulsed ESR experimental capabilities including dynamical decoupling and electron-nuclear double resonance are demonstrated using free radicals diluted in a glassy matrix.

cond-mat.mes-hall↗

arXiv2023

PeaCoK: Persona Commonsense Knowledge for Consistent and Engaging Narratives

Sustaining coherent and engaging narratives requires dialogue or storytelling agents to understand how the personas of speakers or listeners ground the narrative. Specifically, these agents must infer personas of their listeners to produce statements that cater to their interests. They must also learn to maintain consistent speaker personas for themselves throughout the narrative, so that their counterparts feel involved in a realistic conversation or story. However, personas are diverse and complex: they entail large quantities of rich interconnected world knowledge that is challenging to robustly represent in general narrative systems (e.g., a singer is good at singing, and may have attended conservatoire). In this work, we construct a new large-scale persona commonsense knowledge graph, PeaCoK, containing ~100K human-validated persona facts. Our knowledge graph schematizes five dimensions of persona knowledge identified in previous studies of human interactive behaviours, and distils facts in this schema from both existing commonsense knowledge graphs and large-scale pretrained language models. Our analysis indicates that PeaCoK contains rich and precise world persona inferences that help downstream systems generate more consistent and engaging narratives.

cs.CL↗

arXiv2022

Are you aware of what you are watching? Role of machine heuristic in online content recommendations

Since recommender systems have been created and developed to automate the recommendation process, users can easily consume their desired video content on online platforms. In this line, several content recommendation algorithms are introduced and employed to allow users to encounter content of their interests, effectively. However, the recommendation systems sometimes regrettably recommend inappropriate content, including misinformation or fake news. To make it worse, people would unreservedly accept such content due to their cognitive heuristic, machine heuristic, which is the rule of thumb that machines are more accurate and trustworthy than humans. In this study, we designed and conducted a web-based experiment where the participants are invoked machine heuristic by experiencing the whole process of machine or human recommendation system. The results demonstrated that participants (N = 89) showed a more positive attitude toward a machine recommender than a human recommender, even the recommended videos contain inappropriate content. While participants who have a high level of trust in machines exhibited a negative attitude toward recommendations. Based on these results, we suggest that a phenomenon known as algorithm aversion might be simultaneously considered with machine heuristic in investigating human interaction with a machine.

cs.HC↗