arXiv subjects

Jihwan Oh

Publications and source records attributed to Jihwan Oh.

At least 19 recordsLinked to original sources

arXiv2026

Rethinking Personalized Reward Modeling for LLMs under Preference Heterogeneity via Group-Debiased Federated Learning

Large language models are increasingly aligned to human preferences via reward modeling, but user preference data are sensitive and often cannot be centralized. Federated learning keeps such data local while learning a shared initial reward model, which is later personalized for each client through local fine-tuning. Because users often assign opposite labels to the same pair of responses, existing federated methods address preference heterogeneity by clustering similar clients and training one reward model per group, assuming that each group requires its own initialization. We show that this assumption is unnecessary. Under balanced preference groups, a single FedAvg model, despite starting at nearly random accuracy, surpasses reward models trained separately for each ground-truth group after only a few local optimization steps. We attribute this phenomenon to the flatness of the shared initialization: averaging across all clients learns richer shared representations that distinguish responses while canceling conflicting preference directions, leaving the model near a decision boundary that can be rapidly adapted. Group imbalance breaks this effect as the cancellation becomes asymmetric and leaves minority clients too far from the boundary to recover. Motivated by this observation, we propose FedGD (Federated Learning with Group Debiasing), which discovers latent preference groups during federated training and learns a single reward model using group-debiased client sampling. By counteracting the effect of group imbalance, FedGD learns an initialization that remains highly adaptable, enabling effective personalization without prior knowledge of the underlying groups.

cs.LG↗

arXiv2026

PerMix-RLVR: Preserving Persona Expressivity under Verifiable-Reward Alignment

Persona prompting has been widely adopted to steer large language models (LLMs) behavior and improve their instruction performance by assigning specific characters. However, identifying an optimal persona is time-consuming, and its impact on output quality remains poorly understood. Prior work has mainly addressed this issue at the prompt level via inference-time strategies, incurring additional computation. In this work, we avoid inference-time prompt search by tackling persona sensitivity during training, aiming to train models that adapt their behavior to diverse personas while preserving task performance. In particular, we find that reinforcement learning with verifiable rewards (RLVR) systematically reduces sensitivity to persona prompts, but also reveals an inherent trade-off of outcome-based optimization: while RLVR improves robustness on tasks with verifiable goals, it can also degrade persona expressivity when needed, e.g., in-character role-playing. To address this limitation, we propose PerMix-RLVR, a persona-mixed RLVR strategy that mitigates the persona robustness-fidelity trade-off, preserving strong robustness to harmful persona variation while enabling faithful persona adoption when required. Concretely, PerMix-RLVR improves persona stability score (PSS) over RLVR by +21.2% on MATH500, while also enhancing persona fidelity by +11.4% on PersonaGym.

cs.CL↗

arXiv2026

MERIT Feedback Elicits Better Bargaining in LLM Negotiators

Bargaining is often regarded as a logical arena rather than an art or a matter of intuition, yet Large Language Models (LLMs) still struggle to navigate it due to limited strategic depth and difficulty adapting to complex human factors. Current benchmarks rarely capture this limitation. To bridge this gap, we present a utility feedback centric framework. Our contributions are: (i) AgoraBench, a new benchmark spanning nine challenging settings (e.g., deception, monopoly) that supports diverse strategy modeling; (ii) human-aligned, economically grounded metrics derived from utility theory. This is operationalized via agent utility, negotiation power, and acquisition ratio that implicitly measure how well the negotiation aligns with human preference and (iii) a human preference grounded dataset with learning pipeline that strengthens LLMs' bargaining ability through both prompting and finetuning. Empirical results indicate that baseline LLM strategies often diverge from human preferences, while our mechanism substantially improves negotiation performance, yielding deeper strategic behavior and stronger opponent awareness.

cs.AI↗

arXiv2025

Characterizing Compute-Communication Overlap in GPU-Accelerated Distributed Deep Learning: Performance and Power Implications

This paper provides an in-depth characterization of GPU-accelerated systems, to understand the interplay between overlapping computation and communication which is commonly employed in distributed training settings. Due to the large size of models, distributing them across multiple devices is required. Overlapping strategies, which enable concurrent computation and communication, are critical for mitigating communication bottlenecks and maximizing GPU utilization. However, the current consensus is that we should always and aggressively overlap compute and communication to mitigate the overhead of distribution. By systematically evaluating state-of-the-art GPUs, this study investigates the impact of hardware features such as numeric precision, specialized cores, and power capping on distributed training workloads. Comprehensive experiments and studies showcase the effects of overlapping strategies on performance and power consumption across varying scenarios. We observe that overlapping computation and communication can result in an average computational slowdown of 18.9%, with a maximum of 40.0% slowdown. This slowdown is in comparison to the scenario when no communication was happening with the compute. We consider this an ideal execution scenario, where the communication in parallel has not impact on the compute time. However, performing computation and communication sequentially is, on average, 10.2% slower than overlapped execution, with a maximum slowdown of 26.6%. We further observe, while specialized datapath and optimized numeric precision mitigate certain slowdowns, overlapping execution can lead to resource contention and also increase power consumption under specific configurations. The analysis also uncovers trade-offs introduced by power and frequency capping, emphasizing the importance of balanced strategies to optimize energy efficiency and training throughput.

cs.DC↗

arXiv2025

LLM Agents for Bargaining with Utility-based Feedback

Bargaining, a critical aspect of real-world interactions, presents challenges for large language models (LLMs) due to limitations in strategic depth and adaptation to complex human factors. Existing benchmarks often fail to capture this real-world complexity. To address this and enhance LLM capabilities in realistic bargaining, we introduce a comprehensive framework centered on utility-based feedback. Our contributions are threefold: (1) BargainArena, a novel benchmark dataset with six intricate scenarios (e.g., deceptive practices, monopolies) to facilitate diverse strategy modeling; (2) human-aligned, economically-grounded evaluation metrics inspired by utility theory, incorporating agent utility and negotiation power, which implicitly reflect and promote opponent-aware reasoning (OAR); and (3) a structured feedback mechanism enabling LLMs to iteratively refine their bargaining strategies. This mechanism can positively collaborate with in-context learning (ICL) prompts, including those explicitly designed to foster OAR. Experimental results show that LLMs often exhibit negotiation strategies misaligned with human preferences, and that our structured feedback mechanism significantly improves their performance, yielding deeper strategic and opponent-aware reasoning.

cs.LG↗

arXiv2025

ReFeed: Multi-dimensional Summarization Refinement with Reflective Reasoning on Feedback

Summarization refinement faces challenges when extending to multi-dimension. In this paper, we introduce ReFeed, a powerful summarization refinement pipeline that enhances multiple dimensions through reflective reasoning on feedback. To achieve this, we release SumFeed-CoT, a large-scale Long-CoT-based dataset optimized for training a lightweight model with reflective reasoning. Our experiments reveal how the number of dimensions, feedback exposure, and reasoning policy influence refinement performance, highlighting reflective reasoning and simultaneously addressing multiple feedback is crucial to mitigate trade-off between dimensions. Furthermore, ReFeed is robust to noisy feedback and feedback order. Lastly, our finding emphasizes that creating data with a proper goal and guideline constitutes a fundamental pillar of effective reasoning. The dataset and model will be released.

cs.CL↗

arXiv2025

From Belief Entrenchment to Robust Reasoning in LLM Agents

Multi-Agent Debate (MAD) has emerged as a promising inference scaling method for Large Language Model (LLM) reasoning. However, it frequently suffers from belief entrenchment, where agents reinforce shared errors rather than correcting them. Going beyond merely identifying this failure, we decompose it into two distinct root causes: (1) the model's biased $\textit{static initial belief}$ and (2) $\textit{homogenized debate dynamics}$ that amplify the majority view regardless of correctness. To address these sequentially, we propose $\textbf{DReaMAD}$ $($$\textbf{D}$iverse $\textbf{Rea}$soning via $\textbf{M}$ulti-$\textbf{A}$gent $\textbf{D}$ebate with Refined Prompt$)$. Our framework first rectifies the static belief via strategic prior knowledge elicitation, then reshapes the debate dynamics by enforcing perspective diversity. Validated on our new $\textit{MetaNIM Arena}$ benchmark, $\textbf{DReaMAD}$ significantly mitigates entrenchment, achieving a +9.5\% accuracy gain over ReAct prompting and a +19.0\% higher win rate than standard MAD.

cs.LG↗

arXiv2024

Learning to Verify Summary Facts with Fine-Grained LLM Feedback

Training automatic summary fact verifiers often faces the challenge of a lack of human-labeled data. In this paper, we explore alternative way of leveraging Large Language Model (LLM) generated feedback to address the inherent limitation of using human-labeled data. We introduce FineSumFact, a large-scale dataset containing fine-grained factual feedback on summaries. We employ 10 distinct LLMs for diverse summary generation and Llama-3-70B-Instruct for feedback. We utilize this dataset to fine-tune the lightweight open-source model Llama-3-8B-Instruct, optimizing resource efficiency while maintaining high performance. Our experimental results reveal that the model trained on extensive LLM-generated datasets surpasses that trained on smaller human-annotated datasets when evaluated using human-generated test sets. Fine-tuning fact verification models with LLM feedback can be more effective and cost-efficient than using human feedback. The dataset is available at https://github.com/DISL-Lab/FineSumFact.

cs.CL↗

arXiv2024

Learning to Summarize from LLM-generated Feedback

Developing effective text summarizers remains a challenge due to issues like hallucinations, key information omissions, and verbosity in LLM-generated summaries. This work explores using LLM-generated feedback to improve summary quality by aligning the summaries with human preferences for faithfulness, completeness, and conciseness. We introduce FeedSum, a large-scale dataset containing multi-dimensional LLM feedback on summaries of varying quality across diverse domains. Our experiments show how feedback quality, dimensionality, and granularity influence preference learning, revealing that high-quality, multi-dimensional, fine-grained feedback significantly improves summary generation. We also compare two methods for using this feedback: supervised fine-tuning and direct preference optimization. Finally, we introduce SummLlama3-8b, a model that outperforms the nearly 10x larger Llama3-70b-instruct in generating human-preferred summaries, demonstrating that smaller models can achieve superior performance with appropriate training. The full dataset and SummLlama3-8B model are available at https://huggingface.co/datasets/DISLab/FeedSum and https://huggingface.co/DISLab/SummLlama3-8B.

cs.CL↗

arXiv2024

Diffusion-based Episodes Augmentation for Offline Multi-Agent Reinforcement Learning

Offline multi-agent reinforcement learning (MARL) is increasingly recognized as crucial for effectively deploying RL algorithms in environments where real-time interaction is impractical, risky, or costly. In the offline setting, learning from a static dataset of past interactions allows for the development of robust and safe policies without the need for live data collection, which can be fraught with challenges. Building on this foundational importance, we present EAQ, Episodes Augmentation guided by Q-total loss, a novel approach for offline MARL framework utilizing diffusion models. EAQ integrates the Q-total function directly into the diffusion model as a guidance to maximize the global returns in an episode, eliminating the need for separate training. Our focus primarily lies on cooperative scenarios, where agents are required to act collectively towards achieving a shared goal-essentially, maximizing global returns. Consequently, we demonstrate that our episodes augmentation in a collaborative manner significantly boosts offline MARL algorithm compared to the original dataset, improving the normalized return by +17.3% and +12.9% for medium and poor behavioral policies in SMAC simulator, respectively.

cs.LG↗

arXiv2024

Preference Alignment with Flow Matching

We present Preference Flow Matching (PFM), a new framework for preference-based reinforcement learning (PbRL) that streamlines the integration of preferences into an arbitrary class of pre-trained models. Existing PbRL methods require fine-tuning pre-trained models, which presents challenges such as scalability, inefficiency, and the need for model modifications, especially with black-box APIs like GPT-4. In contrast, PFM utilizes flow matching techniques to directly learn from preference data, thereby reducing the dependency on extensive fine-tuning of pre-trained models. By leveraging flow-based models, PFM transforms less preferred data into preferred outcomes, and effectively aligns model outputs with human preferences without relying on explicit or implicit reward function estimation, thus avoiding common issues like overfitting in reward models. We provide theoretical insights that support our method's alignment with standard PbRL objectives. Experimental results indicate the practical effectiveness of our method, offering a new direction in aligning a pre-trained model to preference. Our code is available at https://github.com/jadehaus/preference-flow-matching.

cs.LG↗

arXiv2023

Toward Risk-based Optimistic Exploration for Cooperative Multi-Agent Reinforcement Learning

The multi-agent setting is intricate and unpredictable since the behaviors of multiple agents influence one another. To address this environmental uncertainty, distributional reinforcement learning algorithms that incorporate uncertainty via distributional output have been integrated with multi-agent reinforcement learning (MARL) methods, achieving state-of-the-art performance. However, distributional MARL algorithms still rely on the traditional $ε$-greedy, which does not take cooperative strategy into account. In this paper, we present a risk-based exploration that leads to collaboratively optimistic behavior by shifting the sampling region of distribution. Initially, we take expectations from the upper quantiles of state-action values for exploration, which are optimistic actions, and gradually shift the sampling region of quantiles to the full distribution for exploitation. By ensuring that each agent is exposed to the same level of risk, we can force them to take cooperatively optimistic actions. Our method shows remarkable performance in multi-agent settings requiring cooperative exploration based on quantile regression appropriately controlling the level of risk.

cs.LG↗

arXiv2022

Evidence for an algebra of $G_2$ instantons II

Building on our previous work [2109.01110], we will compute a new kind of $G_2$ instanton partition function. By doing so, we complete a set of building blocks of the instanton partition function associated with a large class of $G_2$ manifolds.

hep-th↗

arXiv2022

The StarCraft Multi-Agent Challenges+ : Learning of Multi-Stage Tasks and Environmental Factors without Precise Reward Functions

In this paper, we propose a novel benchmark called the StarCraft Multi-Agent Challenges+, where agents learn to perform multi-stage tasks and to use environmental factors without precise reward functions. The previous challenges (SMAC) recognized as a standard benchmark of Multi-Agent Reinforcement Learning are mainly concerned with ensuring that all agents cooperatively eliminate approaching adversaries only through fine manipulation with obvious reward functions. This challenge, on the other hand, is interested in the exploration capability of MARL algorithms to efficiently learn implicit multi-stage tasks and environmental factors as well as micro-control. This study covers both offensive and defensive scenarios. In the offensive scenarios, agents must learn to first find opponents and then eliminate them. The defensive scenarios require agents to use topographic features. For example, agents need to position themselves behind protective structures to make it harder for enemies to attack. We investigate MARL algorithms under SMAC+ and observe that recent approaches work well in similar settings to the previous challenges, but misbehave in offensive scenarios. Additionally, we observe that an enhanced exploration approach has a positive effect on performance but is not able to completely solve all scenarios. This study proposes new directions for future research.

cs.LG↗

arXiv2022

Risk Perspective Exploration in Distributional Reinforcement Learning

Distributional reinforcement learning demonstrates state-of-the-art performance in continuous and discrete control settings with the features of variance and risk, which can be used to explore. However, the exploration method employing the risk property is hard to find, although numerous exploration methods in Distributional RL employ the variance of return distribution per action. In this paper, we present risk scheduling approaches that explore risk levels and optimistic behaviors from a risk perspective. We demonstrate the performance enhancement of the DMIX algorithm using risk scheduling in a multi-agent setting with comprehensive experiments.

cs.LG↗

arXiv2022

Integrative Learning of Structured High-Dimensional Data from Multiple Datasets

Integrative learning of multiple datasets has the potential to mitigate the challenge of small $n$ and large $p$ that is often encountered in analysis of big biomedical data such as genomics data. Detection of weak yet important signals can be enhanced by jointly selecting features for all datasets. However, the set of important features may not always be the same across all datasets. Although some existing integrative learning methods allow heterogeneous sparsity structure where a subset of datasets can have zero coefficients for some selected features, they tend to yield reduced efficiency, reinstating the problem of losing weak important signals. We propose a new integrative learning approach which can not only aggregate important signals well in homogeneous sparsity structure, but also substantially alleviate the problem of losing weak important signals in heterogeneous sparsity structure. Our approach exploits a priori known graphical structure of features and encourages joint selection of features that are connected in the graph. Integrating such prior information over multiple datasets enhances the power, while also accounting for the heterogeneity across datasets. Theoretical properties of the proposed method are investigated. We also demonstrate the limitations of existing approaches and the superiority of our method using a simulation study and analysis of gene expression data from ADNI.

stat.ME↗

arXiv2022

Evidence for an Algebra of $\boldsymbol{G_2}$ Instantons

In this short note, we present some evidence towards the existence of an algebra of BPS $G_2$ instantons. These are instantonic configurations that govern the partition functions of 7d SYM theories on local $G_2$ holonomy manifolds $\mathcal X$. To shed light on such structure, we begin investigating the relation with parent 4d $\mathcal N=1$ theories obtained by geometric engineering M-theory on $\mathcal X$. The main point of this paper is to substantiate the following dream: the holomorphic sector of such theories on multi-centered Taub-NUT spaces gives rise to an algebra whose characters organise the $G_2$ instanton partition function. As a first step towards this program, we argue by string duality that a multitude of geometries $\mathcal X$ exist that are dual to well-known 4d SCFTs arising from D3 branes probes of CY cones: all these models are amenable to analysis along the lines suggested by Dijkgraaf, Gukov, Neitzke and Vafa in the context of topological M-theory. Moreover, we discuss an interesting relation to Costello's twisted M-theory, which arises at local patches, and is a key ingredient in identifying the relevant algebras.

hep-th↗

arXiv2022

The Global Form of Flavor Symmetries and 2-Group Symmetries in 5d SCFTs

2-group symmetries arise when 1-form symmetries and 0-form symmetries of a theory mix with each other under group multiplication. We discover the existence of 2-group symmetries in 5d N=1 abelian gauge theories arising on the (non-extended) Coulomb branch of 5d superconformal field theories (SCFTs), leading us to argue that the UV 5d SCFT itself admits a 2-group symmetry. Furthermore, our analysis determines the global forms of the 0-form flavor symmetry groups of 5d SCFTs, irrespective of whether or not the 5d SCFT admits a 1-form symmetry. As a concrete application of our method, we analyze 2-group symmetries of all 5d SCFTs, which reduce in the IR, after performing mass deformations, to 5d N=1 non-abelian gauge theories with simple, simply connected gauge groups. For rank-1 Seiberg theories, we check that our predictions for the flavor symmetry groups match with the superconformal and ray indices available in the literature. We also comment on the mixed 't Hooft anomaly between 1-form and 0-form symmetries arising in 5d N=1 non-abelian gauge theories and its relation to the 2-groups.

hep-th↗