Searcharxiv⌕ Search

arXiv subjects

Junseok Kim

Publications and source records attributed to Junseok Kim.

15 recordsLinked to original sources
arXiv2026

Personalized Privacy Control in LLMs via Attention Head Intervention

The rise of agentic AI enables LLMs to access diverse user data, raising critical privacy concerns. Prior work on contextual privacy studies whether LLMs regulate information disclosure according to context-dependent norms. However, acceptable disclosure boundaries may vary across users even within the same context. To address this limitation, we introduce \textit{personalized privacy}, which incorporates user-specific disclosure preferences into privacy control. We further present P3Bench~(\textbf{P}ersonalized \textbf{P}rivacy \textbf{P}reservation \textbf{Bench}mark), a novel benchmark extending contextual privacy policies with personalized disclosure policies. Experiments show that prompt-based policies fail to reliably enforce personalized privacy policies, with Qwen2.5-7B and Gemma3-4B showing average policy ignorance ratios of 51.25\% and 74.28\%, respectively. Finally, to address this problem, we propose \textsc{Repair}, a robust inference-time attention head intervention method that adjusts disclosure behavior toward policy-consistent responses. Our method significantly improves adherence to user-specific privacy preferences by reducing cases where the model fails to follow the given policy.

cs.AI↗
arXiv2026

Compositional Transduction with Latent Analogies for Offline Goal-Conditioned Reinforcement Learning

Compositional generalization is essential for reaching unseen goals under novel contextual variations in offline goal-conditioned reinforcement learning (GCRL), where a generalist goal-reaching agent must be learned from limited data. Most prior approaches pursue this via trajectory stitching over temporally contiguous segments, which limits composing behaviors across varying contexts. To overcome this limitation, we formalize analogy transduction as synthesizing new plans by composing task-endogenous analogies with given contexts and propose a novel analogy representation tailored for it. Grounded in our theory, this analogy representation captures what changes under optimal task execution, remains invariant to contextual variations, and is sufficient for optimal goal reaching. We further contend that generalization to unseen analogy-context pairs is a practical obstacle in analogy transduction, and introduce a new approach for offline GCRL that enables analogy transduction beyond seen pairs to unseen combinations. We empirically demonstrate the effectiveness of our approach on OGBench manipulation environments, substantially outperforming prior methods that do not perform analogy transduction. Project page: https://rllab-snu.github.io/projects/CTA/

cs.LG↗
arXiv2026

Persona Switch: Mixing Distinct Perspectives in Decoding Time

Role-play prompting is known to steer the behavior of language models by injecting a persona into the prompt, improving their zero-shot reasoning capabilities. However, such improvements are inconsistent across different tasks or instances. This inconsistency suggests that zero-shot and role-play prompting may offer complementary strengths rather than one being universally superior. Building on this insight, we propose Persona Switch, a novel decoding method that dynamically combines the benefits of both prompting strategies. Our method proceeds step-by-step, selecting the better output between zero-shot and role-play prompting at each step by comparing their output confidence, as measured by the logit gap. Experiments with widely-used LLMs demonstrate that Persona Switch consistently outperforms competitive baselines, achieving up to 5.13% accuracy improvement. Furthermore, we show that output confidence serves as an informative measure for selecting the more reliable output.

cs.CL↗
arXiv2026

Reliability-Aware Adaptive Self-Consistency for Efficient Sampling in LLM Reasoning

Self-Consistency improves reasoning reliability through multi-sample aggregation, but incurs substantial inference cost. Adaptive self-consistency methods mitigate this issue by adjusting the sampling budget; however, they rely on count-based stopping rules that treat all responses equally, often leading to unnecessary sampling. We propose Reliability-Aware Adaptive Self-Consistency (ReASC), which addresses this limitation by reframing adaptive sampling from response counting to evidence sufficiency, leveraging response-level confidence for principled information aggregation. ReASC operates in two stages: a single-sample decision stage that resolves instances confidently answerable from a single response, and a reliability-aware accumulation stage that aggregates responses by jointly leveraging their frequency and confidence. Across five models and four datasets, ReASC consistently achieves the best accuracy-cost trade-off compared to existing baselines, yielding improved inference efficiency across model scales from 3B to 27B parameters. As a concrete example, ReASC reduces inference cost by up to 70\% relative to self-consistency while preserving accuracy on GSM8K using Gemma-3-4B-it.

cs.CL↗
arXiv2025

A non-Hopfian ascending HNN-extension of a finitely presented Hopfian group

We find a non-Hopfian ascending HNN-extension of a finitely presented Hopfian group by providing an explicit construction. This result addresses an analogous question to the one posed by Sapir and Wise, which asks whether there is a non-residually finite ascending HNN-extension of a finitely presented residually finite group. Such an analogy is motivated by Mal'cev's result that every finitely generated residually finite group is Hopfian.

math.GR↗
arXiv2024

Stage-Wise Reward Shaping for Acrobatic Robots: A Constrained Multi-Objective Reinforcement Learning Approach

As the complexity of tasks addressed through reinforcement learning (RL) increases, the definition of reward functions also has become highly complicated. We introduce an RL method aimed at simplifying the reward-shaping process through intuitive strategies. Initially, instead of a single reward function composed of various terms, we define multiple reward and cost functions within a constrained multi-objective RL (CMORL) framework. For tasks involving sequential complex movements, we segment the task into distinct stages and define multiple rewards and costs for each stage. Finally, we introduce a practical CMORL algorithm that maximizes objectives based on these rewards while satisfying constraints defined by the costs. The proposed method has been successfully demonstrated across a variety of acrobatic tasks in both simulation and real-world environments. Additionally, it has been shown to successfully perform tasks compared to existing RL and constrained RL algorithms. Our code is available at https://github.com/rllab-snu/Stage-Wise-CMORL.

cs.RO↗
arXiv2024

Persona is a Double-edged Sword: Mitigating the Negative Impact of Role-playing Prompts in Zero-shot Reasoning Tasks

Recent studies demonstrate that prompting a role-playing persona to an LLM improves reasoning capability. However, assigning an adequate persona is difficult since LLMs are extremely sensitive to assigned prompts; thus, inaccurately defined personas sometimes hinder LLMs and degrade their reasoning capabilities. In this paper, we first investigate the potential negative impact of injecting persona into language models. Furthermore, we propose a novel framework, Jekyll \& Hyde, which ensembles the outcomes of both role-playing and neutral prompts to enhance the robustness of reasoning ability. Specifically, Jekyll \& Hyde predicts an appropriate persona using an LLM when defining the role-playing prompt. Then, Jekyll \& Hyde collects two potential solutions from role-playing and neutral prompts and selects a better solution using the LLM evaluator. The experimental analysis demonstrates that role-playing prompts sometimes distract LLMs, degrading their reasoning abilities in 7 out of 12 datasets in llama3. Meanwhile, Jekyll \& Hyde improve reasoning capabilities by selecting better choices among the potential solutions on twelve widely-used natural language reasoning datasets. In addition, we reveal that assigning LLM-generated personas obtains more stable results than handcrafted personas.

cs.CL↗
arXiv2024

Acylindrical hyperbolicity of outer automorphism groups of right-angled Artin groups

We study the acylindrical hyperbolicity of the outer automorphism group of a right-angled Artin group $A_\Gamma$. When the defining graph $\Gamma$ has no SIL-pair (separating intersection of links), we obtain a necessary and sufficient condition for $\mathrm{Out}(A_\Gamma)$ to be acylindrically hyperbolic. As a corollary, if $\Gamma$ is a random connected graph satisfying a certain probabilistic condition, then $\mathrm{Out}(A_\Gamma)$ is not acylindrically hyperbolic with high probability. When $\Gamma$ has a maximal SIL-pair system, we derive a classification theorem for partial conjugations. Such a classification theorem allows us to show that the acylindrical hyperbolicity of $\mathrm{Out}(A_\Gamma)$ is closely related to the existence of a specific type of partial conjugations.

math.GR↗
arXiv2024

Unplug and Play Language Models: Decomposing Experts in Language Models at Inference Time

Enabled by large-scale text corpora with huge parameters, pre-trained language models operate as multi-task experts using a single model architecture. However, recent studies have revealed that certain neurons play disproportionately important roles in solving specific tasks, suggesting that task-relevant substructures can be isolated and selectively activated for each task. Therefore, we introduce Decomposition of Experts (DoE), a novel framework that dynamically identifies and activates task-specific experts within a language model to reduce inference cost without sacrificing accuracy. We first define a task expert as a set of parameters that significantly influence the performance of a specific task and propose a four-step unplug-and-play process: (1) receiving a user request, (2) identifying the corresponding task expert, (3) performing inference using the expert-localized model, and (4) restoring the original model and waiting for the next task. Using attribution methods and prompt tuning, DoE isolates task-relevant neurons, minimizing computational overhead while maintaining task performance. We assume a setting where a language model receives user requests from five widely used natural language understanding benchmarks, processing one task at a time. In this setup, we demonstrate that DoE achieves up to a x1.73 inference speed-up with a 65% pruning rate, without compromising accuracy. Comparisons with various task expert localization methods reveal that DoE effectively identifies task experts, while ablation studies validate the importance of its components. Additionally, we analyze the effects of batch size, token count, and layer types on inference speed-up, providing practical insights for adopting DoE. The proposed framework is both practical and scalable, applicable to any transformer-based architecture, offering a robust solution for efficient task-specific inference.

cs.CL↗
arXiv2022

RAPID: Contention Resolution-based Random Access using Context ID for IoT

With the increasing number of Internet of Things (IoT) devices, Machine Type Communication (MTC) has become an important use case of the Fifth Generation (5G) communication systems. Since MTC devices are mostly disconnected from Base Station (BS) for power saving, random access procedure is required for devices to transmit data. If many devices try random access simultaneously, preamble collision problem occurs, thus causing latency increase. In an environment where delay-sensitive and delay-tolerant devices coexist, the contention-based random access procedure cannot satisfy latency requirements of delay-sensitive devices. Therefore, we propose RAPID, a novel random access procedure, which is completed through two message exchanges for the delay-sensitive devices. We also develop Access Pattern Analyzer (APA), which estimates traffic characteristics of MTC devices. When UEs, performing RAPID and contention-based random access, coexist, it is important to determine a value which is the number of preambles for RAPID to reduce random access load. Thus, we analyze random access load using a Markov chain model to obtain the optimal number of preambles for RAPID. Simulation results show RAPID achieves 99.999% reliability with 80.8% shorter uplink latency, and also decreases random access load by 30.5% compared with state-of-the-art techniques.

cs.NI↗
arXiv2020

Full-stack Hybrid Beamforming in mmWave 5G Networks

This paper analyzes Hybrid Beamforming (HBF) and Multi-User Multiple-Input Multiple-Output (MU-MIMO) in millimeter wave (mmWave) 5th generation (5G) cellular networks considering the full protocol stack with TCP/IP traffic and MAC scheduling. Prior work on HBF and MU-MIMO has assumed full-buffer transmissions and studied link-level performance. We report non-trivial interactions between the HBF technique, the front-loaded channel estimation pilot scheme in NR, and the constraints of MU-MIMO scheduling. We also report that joint multi-user beamforming design is imperative, in the sense that the MU-MIMO system cannot be fully exploited when implemented as a mere collection of single-user analog beams working in parallel. By addressing these issues, throughput can be dramatically increased in mmWave 5G networks by means of Spatial Division Multiple Access (SDMA).

cs.IT↗
arXiv2020

Hybrid Beamforming in 5G mmWave Networks: a Full-stack Perspective

This paper studies the cross-layer challenges and performance of Hybrid Beamforming (HBF) and Multi-User Multiple-Input Multiple-Output (MU-MIMO) in 5G millimeter wave (mmWave) cellular networks with full-stack TCP/IP traffic and MAC scheduling. While previous research on HBF and MU-MIMO has focused on link-level analysis of full-buffer transmissions, this work reveals the interplay between HBF techniques and the higher layers of the protocol stack. To this aim, prior work on full stack simulation of mmWave cellular network has been extended by including the modeling of MU-MIMO and HBF. Our results reveal novel relations between the networking layers and the HBF MU-MIMO performance in the physical layer. Particularly, throughput can be increased in 5G networks by means of Spatial Division Multiple Access (SDMA). However, in order to achieve such benefits it is necessary to take into account certain trade-offs and the implementation complexity of a full-stack HBF solution.

cs.IT↗
arXiv2018

Simultaneous Recognition of Horizontal and Vertical Text in Natural Images

Recent state-of-the-art scene text recognition methods have primarily focused on horizontal text in images. However, in several Asian countries, including China, large amounts of text in signs, books, and TV commercials are vertically directed. Because the horizontal and vertical texts exhibit different characteristics, developing an algorithm that can simultaneously recognize both types of text in real environments is necessary. To address this problem, we adopted the direction encoding mask (DEM) and selective attention network (SAN) methods based on supervised learning. DEM contains directional information to compensate in cases that lack text direction; therefore, our network is trained using this information to handle the vertical text. The SAN method is designed to work individually for both types of text. To train the network to recognize both types of text and to evaluate the effectiveness of the designed model, we prepared a new synthetic vertical text dataset and collected an actual vertical text dataset (VTD142) from the Web. Using these datasets, we proved that our proposed model can accurately recognize both vertical and horizontal text and can achieve state-of-the-art results in experiments using benchmark datasets, including the street view test (SVT), IIIT-5k, and ICDAR. Although our model is relatively simple as compared to its predecessors, it maintains the accuracy and is trained in an end-to-end manner.

cs.CV↗