Searcharxiv⌕ Search

arXiv subjects

Sangwoo Park

Publications and source records attributed to Sangwoo Park.

At least 19 recordsLinked to original sources
arXiv2026

T-MAP: Red-Teaming LLM Agents with Trajectory-aware Evolutionary Search

While prior red-teaming efforts have focused on eliciting harmful text outputs from large language models (LLMs), such approaches fail to capture agent-specific vulnerabilities that emerge through multi-step tool execution, particularly in rapidly growing ecosystems such as the Model Context Protocol (MCP). To address this gap, we propose a trajectory-aware evolutionary search method, T-MAP, which leverages execution trajectories to guide the discovery of adversarial prompts. Our approach enables the automatic generation of attacks that not only bypass safety guardrails but also reliably realize harmful objectives through actual tool interactions. Empirical evaluations across diverse MCP environments demonstrate that T-MAP substantially outperforms baselines in attack realization rate (ARR) and remains effective against frontier models, including GPT-5.2, Gemini-3-Pro, Qwen3.5, and GLM-5, thereby revealing previously underexplored vulnerabilities in autonomous LLM agents. Code is available at https://github.com/pwnhyo/T-MAP.

cs.CR↗
arXiv2026

Reliable Inference in Edge-Cloud Model Cascades via Conformal Alignment

Edge intelligence enables low-latency inference via compact on-device models, but assuring reliability remains challenging. We study edge-cloud cascades that must preserve conditional coverage: whenever the edge returns a prediction set, it should contain the true label with a user-specified probability, as if produced by the cloud model. We formalize conditional coverage with respect to the cloud predictive distribution, and introduce a conformal alignment-based (CAb) cascading mechanism that certifies this property with user control over the risk level. Our method casts escalation from edge to cloud models as a multiple-hypothesis testing (MHT) problem, tailoring conformal alignment (CA) to select which inputs can be safely handled at the edge. The proposed CAb model cascading method yields statistical guarantees on the average fraction of edge decisions that satisfy cloud-level conditional coverage. The procedure applies to arbitrary edge prediction sets, including variants of conformal prediction (CP), and exposes a tunable trade-off among coverage, deferral rate, and set size. Experiments on CIFAR-100 image classification and the TeleQnA question-answering (QA) benchmark show that the proposed CAb cascade maintains the target conditional coverage for edge predictions while substantially reducing offloading to the cloud and incurring modest increases in prediction-set size.

cs.LG↗
arXiv2026

Test-Time Scalable AI-RAN: Inference Time Allocation for Cell-Free MIMO

Artificial intelligence-enabled radio access networks (AI-RANs) are envisioned to consist of multiple AI-based modules, potentially developed independently by different vendors. In this work, we study AI-RAN-enabled cell-free MIMO systems, with a particular focus on the system implications of modern AI models. Specifically, we focus on the phenomenon of test-time scalability popularized by large language models (LLMs), under which model performance improves as additional computational resources are allocated at testing time. By noting that the optimal amount of additional computational resources for each AI module should in general depend on its interaction with the other modules as well as with the underlying wireless channels, we propose a generic framework that enables optimal resource allocation for each test-time scalable module in cell-free MIMO systems. Experimental results demonstrate the effectiveness of the proposed framework in fully exploiting the potential of test-time scalable AI-RANs in cell-free MIMO systems.

eess.SP↗
arXiv2026

DeRP: An Algorithm for Self-Assembly of Power-Delivery Networks using Recursive Branching in Information-Limited Environments

Delivering sustained power to distributed equipment in unstructured field environments using pre-planned wired networks or battery-based solutions presents significant infrastructure and logistics challenges. This paper presents Dendritic Recursive Pivoting (DeRP), a decentralized framework for multi-target network formation in robot swarms based solely on local communication and bearing-based sensing toward sinks. We envision a system in which robots, acting as a conduit, self-assemble a power network from a common source, forming branches at locally selected pivot points that approximate the Steiner points of Steiner trees to efficiently route to multiple Sinks. This branching operation is performed recursively to enable scalable and adaptive network formation without global planning. The proposed method is evaluated in terms of the total network length and estimated power loss, and is quantitatively compared against global baselines such as the Minimum Spanning Tree and Steiner tree solutions (GeoSteiner), which require complete knowledge of Sink locations. Specifically, we found that the networks formed by DeRP asymptotically form approximately 125\% of the global minimum length while reducing power losses to 65\% relative to Euclidean Steiner trees. In addition, we empirically characterize scaling behavior by measuring simulation completion time as the number of Sinks and robots increases, and find that this scaling was sub-linear for up to 100 sinks. The proposed approach enables resilient, adaptive power delivery in environments where deployment of traditional infrastructure is challenging.

cs.RO↗
arXiv2026

Rethinking Reward Models for Multi-Domain Test-Time Scaling

The reliability of large language models (LLMs) during test-time scaling is often assessed with \emph{external verifiers} or \emph{reward models} that distinguish correct reasoning from flawed logic. Prior work has studied both outcome reward models (ORMs), which assess only the final answer, and process reward models (PRMs), which score intermediate reasoning steps. Although PRMs are often viewed as advantageous due to their finer-grained supervision, much of the supporting evidence comes from math-adjacent settings, and their relative benefits across broader domains remain unclear. We present the first unified evaluation of four reward model variants, discriminative ORM and PRM (dORM, dPRM) and generative ORM and PRM (gORM, gPRM), across 14 diverse domains. Contrary to conventional wisdom, we find that (i) dORM performs on par with dPRM, (ii) gPRM is not competitive, and (iii) overall, gORM is the most robust, yielding significant and consistent gains across every tested domain. We attribute the worse performance of gPRM to the stepwise scoring process, which inherits label noise from LLM-based automatic labeling, leading to difficulties in evaluating long reasoning trajectories, including those involving self-correcting reasoning. Both our theoretical analysis and empirical observations indicate that stepwise aggregation compounds errors as reasoning length increases. These findings challenge the common assumption that fine-grained supervision is always better and support generative outcome verification for multi-domain deployment. Our \href{https://github.com/db-Lee/Multi-RM}{\underline{code}} is publicly available to facilitate future research in multi-domain settings.

cs.AI↗
arXiv2026

OmniRetrieval: Unified Retrieval across Heterogeneous Knowledge Sources

Real-world information needs require access to structurally diverse knowledge sources, from unstructured text and relational tables to knowledge graphs and property graphs. Existing retrievers, however, operate over one source at a time under a fixed query language, leaving the broader landscape of available knowledge fragmented behind incompatible interfaces. A natural attempt at unification would collapse these sources into a shared space, but this erases the structural affordances (such as schemas, ontologies, compositional operators) that give each source its expressive power. Effective retrieval over diverse knowledge, therefore, requires not homogenization but an overarching layer that meets each source on its own terms. To achieve this, we present OmniRetrieval, a framework that takes any natural-language query, identifies appropriate knowledge sources, and dispatches source-native queries to their native execution engines. Across an extensive benchmark spanning 13 datasets and 309 distinct knowledge bases over text, relational, and graph-structured sources, OmniRetrieval exceeds single-source baselines, demonstrating that it can serve as a general-purpose interface to the heterogeneous sources while preserving the structural distinctions that make each source valuable.

cs.CL↗
arXiv2026

Reliable Wireless Indoor Localization via Cross-Validated Prediction-Powered Calibration

Wireless indoor localization using predictive models with received signal strength information (RSSI) requires proper calibration for reliable position estimates. One remedy is to employ synthetic labels produced by a (generally different) predictive model. But fine-tuning an additional predictor, as well as estimating residual bias of the synthetic labels, demands additional data, aggravating calibration data scarcity in wireless environments. This letter proposes an approach that efficiently uses limited calibration data to simultaneously fine-tune a predictor and estimate the bias of synthetic labels, yielding prediction sets with rigorous coverage guarantees. Experiments on a fingerprinting dataset validate the effectiveness of the proposed method.

cs.LG↗
arXiv2026

It Takes Two: Complementary Self-Distillation for Contextual Integrity in LLMs

Contextual Integrity (CI) defines privacy not merely as keeping information hidden, but as governing information flows according to the norms of a given context. As large language models are increasingly deployed as personal agents handling sensitive workflows, adhering to CI becomes critical. However, even frontier models remain unreliable in making disclosure decisions, and existing mitigation strategies often degrade underlying task performance. To overcome this privacy-utility trade-off, we propose SELFCI, a complementary self-distillation framework that decouples information suppression from task resolution. SELFCI jointly optimizes two independent reverse KL divergences over distinct teacher distributions derived from feedback: one encourages preserving task-relevant information for utility, while the other enforces minimal and appropriate disclosure. This complementary formulation induces a Product-of-Experts (PoE) target, aligning the policy with the intersection of capability and privacy requirements. Empirical evaluations demonstrate that SELFCI, without relying on costly external supervision, consistently outperforms competitive baselines such as online reinforcement learning algorithms (e.g., GRPO). These trends further extend to out-of-domain settings involving agentic workflows and accumulated private context, suggesting that SELFCI provides a practical path toward CI alignment.

cs.LG↗
arXiv2026

Nudging Beyond the Comfort Zone: Efficient Strategy-Guided Exploration for RLVR

Reinforcement learning with verifiable rewards (RLVR) has emerged as a scalable paradigm for improving the reasoning capabilities of large language models. However, its effectiveness is fundamentally limited by exploration: the policy can only improve on trajectories it has already sampled. While increasing the number of rollouts alleviates this issue, such brute-force scaling is computationally expensive, and existing approaches that modify the optimization objective provide limited control over what is explored. In this work, we propose NudgeRL, a framework for structured and diversity-driven exploration in RLVR. Our approach introduces Strategy Nudging, which conditions each rollout on lightweight, strategy-level contexts to induce diverse reasoning trajectories without relying on expensive oracle supervision. To effectively learn from such structured exploration, we further propose a unified objective, which decomposes the reward signal into inter- and intra-context components and incorporates a distillation objective to transfer discovered behaviors back to the base policy. Empirically, NudgeRL outperforms standard GRPO with up to 8 times larger rollout budgets, while outperforming oracle-guided RL baseline on average across five challenging math benchmarks. These results demonstrate that structured, context-driven exploration can serve as an efficient and scalable alternative to both brute-force rollout scaling and feasibility-oriented methods based on privileged information. Our code is available at https://github.com/tally0818/NudgeRL.

cs.AI↗
arXiv2026

THINKSAFE: Self-Generated Safety Alignment for Reasoning Models

Large reasoning models (LRMs) achieve remarkable performance by leveraging reinforcement learning (RL) on reasoning tasks to generate long chain-of-thought (CoT) reasoning. However, this over-optimization often prioritizes compliance, making models vulnerable to harmful prompts. To mitigate this safety degradation, recent approaches rely on external teacher distillation, yet this introduces a distributional discrepancy that degrades native reasoning. We formalize safety realignment as a KL projection onto the safe simplex and prove that the student's own safety-filtered distribution is the unique KL-optimal target, while any external teacher incurs an irreducible excess KL penalty. Guided by this analysis, we propose ThinkSafe, a self-generated alignment framework that restores safety without external teachers. Our key insight is that while compliance suppresses safety mechanisms, models often retain latent knowledge to identify harm. ThinkSafe unlocks this via lightweight refusal steering, which preserves the KL-optimal target while increasing the acceptance rate. Experiments on DeepSeek-R1-Distill and Qwen3 show ThinkSafe significantly improves safety while preserving reasoning proficiency, and achieves superior safety and comparable reasoning to GRPO with roughly an order of magnitude less compute. Code, models, and datasets are available at https://github.com/seanie12/ThinkSafe and https://huggingface.co/Seanie-lee/collections.

cs.AI↗
arXiv2026

PREPING: Building Agent Memory without Tasks

Agent memory is typically constructed either offline from curated demonstrations or online from post-deployment interactions. However, regardless of how it is built, an agent faces a cold-start gap when first introduced to a new environment without any task-specific experience available. In this paper, we study pre-task memory construction: whether an agent can build procedural memory before observing any target-environment tasks, using only self-generated synthetic practice. Yet, synthetic interaction alone is insufficient, as without controlling what to practice and what to store, synthetic tasks become redundant, infeasible, and ultimately uninformative, and memory further degrades quickly due to unfiltered trajectories. To overcome this, we present Preping, a proposer-guided memory construction framework. At its core is proposer memory, a structured control state that shapes future practice. A Proposer generates synthetic tasks conditioned on this state, a Solver executes them, and a Validator determines which trajectories are eligible for memory insertion while also providing feedback to guide future proposals. Experiments on AppWorld, BFCL v3, and MCP-Universe show that Preping substantially improves over a no-memory baseline and achieves performance competitive with strong playbook-based methods built from offline or online experience, with deployment cost $2.99\times$ lower on AppWorld and $2.23\times$ lower on BFCL v3 than online memory construction. Further analyses reveal that the main benefit does not come from synthetic volume alone, but from proposer-side control over feasibility, redundancy, and coverage, combined with selective memory updates.

cs.AI↗
arXiv2026

Reliable LLM-Based Edge-Cloud-Expert Cascades for Telecom Knowledge Systems

Large language models (LLMs) are emerging as key enablers of automation in domains such as telecommunications, assisting with tasks including troubleshooting, standards interpretation, and network optimization. However, their deployment in practice must balance inference cost, latency, and reliability. In this work, we study an edge-cloud-expert cascaded LLM-based knowledge system that supports decision-making through a question-and-answer pipeline. In it, an efficient edge model handles routine queries, a more capable cloud model addresses complex cases, and human experts are involved only when necessary. We define a misalignment-cost constrained optimization problem, aiming to minimize average processing cost, while guaranteeing alignment of automated answers with expert judgments. We propose a statistically rigorous threshold selection method based on multiple hypothesis testing (MHT) for a query processing mechanism based on knowledge and confidence tests. The approach provides finite-sample guarantees on misalignment risk. Experiments on the TeleQnA dataset -- a telecom-specific benchmark -- demonstrate that the proposed method achieves superior cost-efficiency compared to conventional cascaded baselines, while ensuring reliability at prescribed confidence levels.

eess.SP↗
arXiv2026

Test-Beam Performance of the AstroPix Silicon Sensor for Imaging Calorimetry

AstroPix is a high-voltage CMOS HVCMOS monolithic active pixel sensor MAPS developed for future space-based gamma-ray missions. It is also a candidate technology for the imaging layer of the Barrel Imaging Calorimeter BIC in the ePIC experiment at the future Electron-Ion Collider EIC. We report the first AstroPix test-beam results obtained at the KEK Photon Factory Advanced Ring PF-AR and the CERN Proton Synchrotron PS T10 beam line in 2025, using the third prototype AstroPix-v3. AstroPix-v3 sensors were operated as both standalone tracking layers and imaging layers interleaved with prototype lead/scintillating-fiber Pb/SciFi calorimeter modules, using electron and hadron beams in the few-GeV/c momentum range. Event synchronization between the continuous readout of AstroPix-v3 and the trigger-based readout of the Pb/SciFi calorimeter was achieved using a common timestamp. The AstroPix-v3 sensors exhibit stable performance, reaching a maximum hit efficiency of 68 percent at a bias voltage of -400 V under pion-dominated beam conditions. When combined with the Pb/SciFi calorimeter, the AstroPix layers successfully capture the development of electromagnetic showers. Using Cherenkov-based particle identification, electron-induced events exhibit significantly higher hit multiplicities and broader spatial distributions than pion-induced events, thereby providing clear discrimination between electromagnetic and hadronic showers. These results demonstrate that AstroPix-v3 provides effective, high-granularity imaging of shower development and is well suited as an imaging layer in future calorimeter systems for both collider and space-based experiments.

physics.ins-det↗
arXiv2026

Adaptive Selection of LoRA Components in Privacy-Preserving Federated Learning

Differentially private federated fine-tuning of large models with LoRA suffers from aggregation error caused by LoRA's multiplicative structure, which is further amplified by DP noise and degrades both stability and accuracy. Existing remedies apply a single update mode uniformly across all layers and all communication rounds (or alternate them on a fixed schedule), ignoring both the structural asymmetry between the two LoRA factors and the round-wise dynamics of training. We propose AS-LoRA, an adaptive framework defined by three axes (i) layer-wise freedom, in which each layer independently selects its active component, (ii) round-wise adaptivity, in which the selection updates over communication rounds, and (iii) a curvature-aware score derived from a second-order approximation of the loss. Theoretically, AS-LoRA eliminates the reconstruction-error floor of layer-tied schedules, accelerates convergence, implicitly biases solutions toward flatter minima, and incurs no additional privacy cost. Across GLUE, SQuAD, CIFAR-100, and Tiny-ImageNet under strict DP budgets and non-IID partitions, AS-LoRA improves over the federated LoRA baselines by up to $+7.5$ pp on GLUE and $+12.5$ pp on MNLI-mm for example, while matching or exceeding SVD-based aggregation methods at $33\text{--}180 \times$ lower aggregation cost and with negligible communication overhead. Code for the proposed method is available at https://anonymous.4open.science/r/as_lora-F75F/.

cs.LG↗
arXiv2026

On-Device Fine-Tuning via Backprop-Free Zeroth-Order Optimization

On-device fine-tuning is a critical capability for edge AI systems, which must support adaptation to different agentic tasks under stringent memory constraints. Conventional backpropagation (BP)-based training requires storing layer activations and optimizer states, a demand that can be only partially alleviated through checkpointing. In edge deployments in which the model weights must reside entirely in device memory, this overhead severely limits the maximum model size that can be deployed. Memory-efficient zeroth-order optimization (MeZO) alleviates this bottleneck by estimating gradients using forward evaluations alone, eliminating the need for storing intermediate activations or optimizer states. This enables significantly larger models to fit within on-chip memory, albeit at the cost of potentially longer fine-tuning wall-clock time. This paper first provides a theoretical estimate of the relative model sizes that can be accommodated under BP and MeZO training. We then numerically validate the analysis, demonstrating that MeZO exhibits accuracy advantages under on-device memory constraints, provided sufficient wall-clock time is available for fine-tuning.

cs.LG↗
arXiv2026

Finding Strongly Lensed Supernovae from Blended Light Curves

We present a model-independent, photometry-only framework for identifying strongly lensed supernovae when multiple images are unresolved and blended into a single point source. Building on the simulation-based methodology of Bag et al. (2021), we apply this approach to real Zwicky Transient Facility (ZTF) data using a validation sample of spectroscopically confirmed Type Ia supernovae. The method models the observed flux as a superposition of two time-shifted components, and Bayesian inference is used to estimate the relative scaling and time delay. Applying this framework to 445 well-converged supernovae, we find that only a single object satisfies the selection criteria when adopting a conservative threshold of $Δt \ge 12$ days, corresponding to a false positive fraction of $1/445 \approx 0.22\%$. A laxer threshold of $Δt \ge 10$ days yields fourteen objects, for a false positive fraction of $3.15\%$. The method provides a scalable and model-independent first-stage filter for identifying lens-like candidates in large time-domain surveys such as the Rubin Observatory's Legacy Survey of Space and Time (LSST).

astro-ph.IM↗
arXiv2026

Beam test of a Pb/SciFi prototype for the Barrel Imaging Calorimeter at the Electron-Ion Collider

A Lead-Scintillating Fiber (Pb/SciFi) prototype for the Barrel Imaging Calorimeter (BIC) at the Electron--Ion Collider (EIC) was tested with electron beams at the CERN PS T10 beam line in August 2024. The prototype consisted of unit modules with a sampling structure of lead sheets and scintillating fibers, corresponding to a total depth of approximately $10.9\,X_{0}$. Beam tests were performed with electron momenta between 0.5 and 3~GeV/$c$ to evaluate the energy and timing performance of the prototype. This study characterizes the performance of a Pb/SciFi prototype and provides input for future beam tests, calibration and readout optimization, and the development of larger-scale prototypes.

physics.ins-det↗
arXiv2026

Chain of Retrieval: Multi-Aspect Iterative Search Expansion and Post-Order Search Aggregation for Full Paper Retrieval

Scientific paper retrieval, particularly framed as document-to-document retrieval, aims to identify relevant papers in response to a long-form query paper, rather than a short query string. Previous approaches to this task have focused exclusively on abstracts, embedding them into dense vectors as surrogates for full documents and calculating similarity between them. Yet, abstracts offer only sparse and high-level summaries, and such methods primarily optimize one-to-one similarity, overlooking the dynamic relations that emerge across relevant papers during the retrieval process. To address this, we propose Chain of Retrieval(COR), a novel iterative framework for full-paper retrieval. Specifically, COR decomposes each query paper into multiple aspect-specific views, matches them against segmented candidate papers, and iteratively expands the search by promoting top-ranked results as new queries, thereby forming a tree-structured retrieval process. The resulting retrieval tree is then aggregated in a post-order manner: descendants are first combined at the query level, then recursively merged with their parent nodes, to capture hierarchical relations across iterations. To validate this, we present SCIFULLBENCH, a large-scale benchmark providing both complete and segmented contexts of full papers for queries and candidates, and results show that COR significantly outperforms existing retrieval baselines. Our code and dataset is available at https://github.com/psw0021/Chain-of-Retrieval-Official.

cs.IR↗