Searcharxiv⌕ Search

arXiv subjects

Hyunjun Kim

Publications and source records attributed to Hyunjun Kim.

At least 19 recordsLinked to original sources

CyBiasBench: Benchmarking Bias in LLM Agents for Cyber-Attack Scenarios

Large language models (LLMs) are increasingly deployed as autonomous agents in offensive cybersecurity. In this paper, we reveal an interesting phenomenon: different agents exhibit distinct attack patterns. Specifically, agents exhibit distinct attack-selection biases, differing in both the families they explore and the concentration of their efforts. These patterns remain relatively stable across prompt variations. To systematically quantify this behavior, we introduce CyBiasBench, a comprehensive 630-session benchmark that evaluates five agents on three targets and four prompt conditions with ten attack families. We identify explicit bias across agents, with different dominant attack families and varying entropy levels in their attack-family allocation distributions. Such bias characterizes agent behavior within the evaluated environments and provides information beyond aggregate attack success rates. Furthermore, our experiments reveal a bias momentum effect, where agents resist explicit steering toward attack families that conflict with their bias. This forced distribution shift does not yield measurable improvements in attack performance. To ensure reproducibility and facilitate future research, we release an interactive result dashboard at https://trustworthyai.co.kr/CyBiasBench/ and a reproducibility artifact with aggregated session-level statistics and full evaluation scripts at https://github.com/Harry24k/CyBiasBench.

cs.CR↗

MiniPose: Miniature-Based Real Data Collection for 6D Pose Estimation of Large Industrial Pipe Segments

6-DOF pose estimation of large and heavy industrial objects, such as large piping systems in shipyards and structural components in petrochemical plants, remains relatively underexplored in robotic perception. Repeatedly repositioning large objects is physically and spatially demanding, making it difficult to acquire real training data with sufficient pose diversity. Synthetic data offer a scalable alternative, but faithfully reproducing real RGB-D observations, particularly the structured measurement errors of real depth sensors, remains difficult. We propose MiniPose, a practical pipeline for miniature-based real RGB-D data collection that addresses both the data-acquisition burden and the sim-to-real gap. We fabricate a scaled-down replica and capture RGB-D data of it in diverse poses using a real RGB-D camera. RGB images are used directly, while depth maps are converted to a geometrically consistent representation of the full-scale object via a simple inverse scaling transformation. This allows training to incorporate real appearance and sensor measurement characteristics that are difficult to reproduce using synthetic data alone. To examine whether the benefit depends on a particular network, we evaluate two RGB-D pose estimators with different formulations, PVN3D and RDPN6D. Mixing at most 1,000 miniature images per pipe type, less than 10\% of the training data, with 10,000 synthetic images improves the mean ADD-S AUC across pipe types over synthetic-only training for both estimators. A comparison with training on full-scale real data further suggests that miniature data can serve as a proxy where full-scale collection is difficult, reducing the data-collection burden for large industrial objects.

cs.RO↗

Agents' Last Exam

Recent AI systems have achieved strong results on a wide range of benchmarks, yet these gains have not translated into economically meaningful deployment across many professional domains. We argue that this gap is largely an evaluation problem: widely used benchmarks lack sustained performance measurement on real and economically valuable workflows. This paper introduces Agents' Last Exam (ALE), a benchmark designed to evaluate AI agents on long horizon, economically valuable, real world tasks with verifiable outcomes. Developed in collaboration with 250+ industry experts, ALE covers non-physical industries defined with reference to O*NET / SOC 2018 (the U.S. federal occupational taxonomy). It is organized around a task taxonomy with 55 sub fields grouped into 13 industry clusters covering 1K+ tasks. Current results show that the hardest tier remains far from saturated: across mainstream harness and backbone configurations, the average full pass rate is below 1%. ALE is designed as a living benchmark: its task pool grows continuously as new workflows and industries are onboarded. More broadly, ALE is intended not merely as another leaderboard, but as an instrument for closing the gap between benchmark success and GDP relevant impact.

cs.AI↗

DeepAVR: Deep Analysis via Multi-Agent Collaboration for Automated Vulnerability Repair

Automated Vulnerability Repair (AVR) has emerged as a critical defense against software security flaws, aiming to triage and fix bugs autonomously. Early LLM-based approaches focused on patch generation under the assumption of perfect fault localization, while recent end-to-end methods attempt to repair vulnerabilities directly from sanitizer logs. However, these methods often fail when the root cause is deeply hidden--either physically distant from the crash site or logically masked by interdependent errors--making the initial report insufficient. We propose DeepAVR, a multi-agent framework for end-to-end AVR that performs deep analysis by transforming intermediate repair outcomes into structured diagnostic feedback. DeepAVR decomposes repair into specialized roles--Fault Localization, Patch Generation, Compile Validation, Security Validation, Functionality Validation, and Patch Review--and routes feedback across these roles to iteratively refine both localization and patch hypotheses. In particular, DeepAVR introduces patch-conditioned re-localization, which treats execution results from intermediate patches as diagnostic signals to uncover hidden root causes, and a root-cause-oriented patch review stage that detects symptom-level fixes and incomplete repairs beyond execution-based validation. Evaluation on the real-world C/C++ vulnerability repair dataset VulnLoc shows that DeepAVR significantly outperforms prior methods, generating correct patches for 92.3% of vulnerabilities. Furthermore, results on SAN2VULN demonstrate strong generalization, with DeepAVR effectively repairing vulnerabilities unseen during training.

cs.SE↗

Engram Memory Network: Brain-Inspired Prototype Explanations

Artificial intelligence systems are increasingly deployed in high-stakes decision-making. In response, prototype-based models in explainable AI have attracted growing attention for their ability to retrieve visually similar examples as faithful explanations. Although these models achieve competitive classification accuracy, they often exhibit degraded explanation quality across datasets with varying characteristics, limiting their practical applicability. We draw on insights from cognitive neuroscience, particularly the human memory system, which encodes input-relevant information into representations that support recognition. We propose an Engram Memory Network (EMN), a model that emulates the ventral visual stream and inferior temporal cortex in the human brain. Central to this model is a Hopfield-inspired memory module that iteratively retrieves representative prototypes. These memory-refined latents are used for both reconstruction and classification, encouraging the selected prototypes to capture input-relevant information across diverse datasets. To address the lack of standardized evaluation for prototype explanations, we assess three complementary aspects: perceptual similarity (LPIPS and DISTS) for perceived resemblance, subclass alignment, measuring whether prototypes match the input's fine-grained category beyond supervised labels, and information preservation (NMI), measuring how much input information is captured by selected prototypes. Experiments show that EMN improves visual similarity by 17.06%, subclass alignment by 104.57%, and NMI by 42.42% on average across CIFAR-10/100, MNIST, HAM10000, and CUB-200, while achieving the highest average classification accuracy (87.62%) among all methods. These findings underscore the effectiveness of neuroscience-inspired explanation mechanisms in advancing the interpretability and reliability of prototype-based models.

cs.LG↗

Causal limits and optimal allocation for passive frequency-selective hearing protection

Passive Helmholtz-resonator earplugs aim to attenuate hazardous noise while keeping speech audible; the air volume of an ear-sized shell is finite, and we treat it as the resource a design spends. For a vented bore transparent at zero frequency and loaded by reactive side branches, causality fixes a conserved integral of transmission loss. We derive it in device variables, add viscous and flanking constraints, and allocate the budget by hazard-weighted water-filling with a continuous speech-cost price. The audio-band budget B_ad = (20/ln10) pi^2 V / 2cS is set by the ratio of total cavity volume V to bore area S; occluding plugs are exempt. The f^2 price signal favours frequencies well above the source peak. Replacing strict transparency by a speech-cost price gives an exchange rate of about 1.3 dB of protection per decibel of speech cost under the present model, with the budget setting where it saturates. Across three form factors, low-frequency-selective designs stay near a decibel while hazard-matched mid/high-frequency designs reach twelve. The binding variables are bore area and speech allowance, not resonator count.

cs.SD↗

Lesion-centered 3D mapping of colonoscopy procedures: validation of a hierarchical ensemble pipeline on public benchmark videos

Background and Objective: Colonoscopy recording practice preserves text reports and still photographs, while the spatial information already present in the recorded video - where the scope traveled, where a lesion was observed, and whether the same lesion was seen again - is discarded when the procedure ends. This study determines whether a lesion-centered spatial record can be assembled and validated without full-colon 3D reconstruction. Methods: A four-layer hierarchical pipeline was assembled - (1) a global topological map, (2) lesion-level spatio-temporal tracks, (3) on-demand local 3D reconstruction, and (4) persistent lesion identity across repeated observations - and ran end to end on four public videos (two C3VDv2 sequences with ground-truth depth and two full REAL-Colon procedures; 40,245 frames). All components are published, individually validated methods; the contribution is their lesion-centered assembly, linking rules, and evaluation. Results: Revisits, impossible under forward-only mapping by construction, were detected by entry-map Bayesian localization: 5,614 and 4,043 revisit events (56 and 68 distinct nodes) in the two full procedures. Lesion-identity merging at the adopted threshold 0.5 maintained ground-truth purity 1.0 while auto-merging 20 of 231 candidate pairs. The endoscopy-specific geometry engine outperformed a general-purpose foundation model on all metrics (overall absolute relative error (AbsRel) 0.2276 vs. 0.3523). Conclusions: The results are partial but establish a concrete near-term path: revisit detection, lesion identity, and local 3D each returned quantitative, reproducible output without waiting for complete geometric reconstruction; validating the record on clinical data is the next step.

cs.CV↗

Experimental Analysis of Productive Interaction Strategy with ChatGPT: User Study on Function and Project-level Code Generation Tasks

The application of Large Language Models (LLMs) is growing in the productive completion of Software Engineering tasks. Yet, studies investigating productive prompting techniques often employed a limited problem space, focusing primarily on well-known prompting patterns and targeting function-level SE practices. We identify significant gaps in real-world workflows that involve complexities beyond class-level (e.g., multi-class dependencies) and different features that can impact Human-LLM Interaction (HLI) processes in code generation. To address these issues, we designed an experiment to comprehensively analyze HLI features related to code generation productivity. Our study presents two project-level benchmark tasks that extend beyond function-level evaluations. We conducted a user study with 36 participants from diverse backgrounds, asking them to solve the assigned tasks by interacting with the GPT assistant using specific prompting patterns. We also examined participants' experiences and behavioral features during interactions by analyzing screen recordings and GPT chat logs. Our empirical investigation, based on statistical guidance, revealed (1) that three out of 15 HLI features emerged as consistently supported factors for productivity; (2) five primary guidelines for enhancing productivity for HLI processes; and (3) a taxonomy of 29 runtime and logic errors that can occur during HLI processes, along with suggested mitigation plans.

cs.SE↗

EquiGQNet: Fast Grasp Quality Evaluation via Shared Equivariant Point Cloud Encoding

Planning six-degree-of-freedom (6-DoF) grasps for unseen objects in cluttered tabletop scenes from a single-view depth image requires accurate and efficient evaluation of diverse grasp candidates. Existing early-fusion methods capture local object geometry relative to each grasp candidate but repeatedly encode the scene, whereas late-fusion methods reuse a shared scene representation but may lose this grasp-relative local geometry. We propose EquiGQNet, an efficient 6-DoF grasp quality evaluator that combines the strengths of both approaches. For grasp orientation, EquiGQNet replaces the early-fusion operation of rotating and re-encoding the point cloud for each grasp candidate with an SO(3)-equivariant encode-once-then-rotate scheme, yielding grasp-aligned geometric features from a shared scene encoding. For grasp translation, Mid-level Action Fusion (MAF) injects the grasp position into intermediate features before global aggregation, retaining local geometry relative to each candidate. We evaluate EquiGQNet in two grasp planning pipelines: Cross-Entropy Method (CEM)-based continuous grasp search and candidate ranking with a pretrained generative planner. In simulation, EquiGQNet achieves grasping performance comparable to the early-fusion baseline and substantially outperforms late fusion on objects with complex geometry and limited graspable regions, while reducing CEM planning time from 3.31s to 0.48s, a 6.9x speedup over early fusion. In real-world household-object decluttering, EquiGQNet achieves a 95.2% grasp success rate and 230 picks per hour, versus 153 and 170 for early- and late-fusion baselines. Code is available at https://equigqnet.github.io/.

cs.RO↗

URF: A Unified Robot Control-Policy Framework for Stable Contact Aware Manipulation

Learning-based manipulation policies usually predict robot actions from sensory observations and leave their execution to a separate low-level controller. In rigid contact, this separation can be problematic: the same motion to a virtual target or compliant motion command can lead to unstable contact, tracking error, excessive loading, or tool damage, depending on the low-level controller. In this paper, we propose a \textit{Unified Robot Control-Policy Framework} (URF), which connects compliant action prediction with unified impedance-admittance control. Given multimodal observations, URF predicts a virtual target, a stiffness matrix, and an impedance-admittance switch ratio. The switch ratio determines when the controller should behave more like admittance control for accurate motion tracking and when it should move toward impedance control for safer rigid contact. Because demonstration data do not provide ground-truth environment stiffness, we construct switch-ratio labels from measured contact forces and use them to supervise controller-mode prediction. Across box-flipping and line-pressing tasks, URF achieves higher task success rates while reducing failure modes observed with admittance-only execution, including rapid force buildup, large force oscillations, tool breakage, and robot safety stops. These results suggest that contact-aware policies benefit from predicting not only compliant actions but also the controller behavior used to execute them. Project page: https://jiyou384.github.io/urf_project_page/

cs.RO↗

Technical Report: AI-Assisted Gated DeltaNet Optimization on NVIDIA Blackwell

AI-assisted GPU programming is often framed as a kernel-generation loop: ask a model to produce faster CUDA code, benchmark the result, and repeat. This case study argues that contest-grade optimization involves more than improving the kernel body. We examine the Agent-Assisted submission by our team, MSInfer, to the MLSys 2026 FlashInfer Contest. The submission optimized Gated DeltaNet decode and prefill on NVIDIA B200/Blackwell and achieved an official $1.58\times$ speedup, with approximate average latencies of $9.315\,μ\mathrm{s}$ for decode and $239.48\,μ\mathrm{s}$ for prefill. Our experience shows that even effective local kernel improvements can plateau when a workload requires structural reformulation and evaluator-aligned measurement. We therefore characterize AI-assisted kernel optimization as an end-to-end systems problem that encompasses algorithm design, workload specialization, measurement tooling, build and evaluation surfaces, evaluator alignment, and human interpretation.

cs.DC↗

ReFoCUS: Reinforcement-guided Frame Optimization for Contextual Understanding

Recent progress in Large Multi-modal Models (LMMs) has enabled effective vision-language reasoning, yet the ability to video understanding remains constrained by suboptimal frame selection strategies, albeit with the rapid development of video-specialized LMMs. Prior works attempted to solve this with static heuristics or external retrieval modules to feed frame-level information, but these approaches often fail to capture visual cues grounded to the given user queries conflating raw visual dynamics with true semantic relevance. In this paper, we introduce ReFoCUS (Reinforcement-guided Frame Optimization for Contextual UnderStanding), the first framework to integrate online policy-gradient reinforcement learning into frame-level optimization for video-LLMs. ReFoCUS aims to learn a frame selection policy, leveraging reward signals derived from reference models to capture their underlying scoring behavior over frame combinations that best support temporally grounded responses. To efficiently explore the large combinatorial frame space, we employ an autoregressive and query-conditional selection architecture that ensures contextual consistency while reducing complexity. Our policy learning removes the need for explicit frame-level supervision, as it implicitly discovers optimal and semantically consistent frame compositions. ReFoCUS consistently improves reasoning accuracy across multiple video QA benchmarks, demonstrating the advantage of aligning frame selection with model-internal utility.

cs.CV↗

Robust Grounding with MLLMs Against Occlusion and Small Objects via Language-Guided Semantic Cues

While Multimodal Large Language Models (MLLMs) have enhanced grounding capabilities in general scenes, their robustness in crowded scenes remains underexplored. Crowded scenes entail visual challenges (i.e., occlusion and small objects), which impair object semantics and degrade grounding performance. In contrast, language expressions are immune to such degradation and preserve object semantics. In light of these observations, we propose a novel method that overcomes such constraints by leveraging Language-Guided Semantic Cues (LGSCs). Specifically, our approach introduces a Semantic Cue Extractor (SCE) to derive semantic cues of objects from the visual pipeline of an MLLM. We then guide these cues using corresponding text embeddings to produce LGSCs as linguistic semantic priors. Subsequently, they are reintegrated into the original visual pipeline to refine object semantics. Extensive experiments and analyses demonstrate that incorporating LGSCs into an MLLM effectively improves grounding accuracy in crowded scenes.

cs.CV↗

Analyzing the Training Dynamics of Image Restoration Transformers: A Revisit to Layer Normalization

This work analyzes the training dynamics of Image Restoration (IR) Transformers and uncovers a critical yet overlooked issue: conventional LayerNorm (LN) drives feature magnitudes to diverge to a million scale and collapses channel-wise entropy. We analyze this in the perspective of networks attempting to bypass LN's constraints that conflict with IR tasks. Accordingly, we address two misalignments between LN and IR: 1) per-token normalization disrupts spatial correlations, and 2) input-independent scaling discards input-specific statistics. To address this, we propose Image Restoration Transformer Tailored Layer Normalization i-LN, a simple drop-in replacement that normalizes features holistically and adaptively rescales them per input. We provide theoretical insights and empirical evidence that this simple design effectively leads to both improved training dynamics and thereby improved performance, validated by extensive experiments.

cs.CV↗

LION-DG: Layer-Informed Initialization with Deep Gradient Protocols for Accelerated Neural Network Training

Weight initialization remains decisive for neural network optimization, yet existing methods are largely layer-agnostic. We study initialization for deeply-supervised architectures with auxiliary classifiers, where untrained auxiliary heads can destabilize early training through gradient interference. We propose LION-DG, a layer-informed initialization that zero-initializes auxiliary classifier heads while applying standard He-initialization to the backbone. We prove that this implements Gradient Awakening: auxiliary gradients are exactly zero at initialization, then phase in naturally as weights grow -- providing an implicit warmup without hyperparameters. Experiments on CIFAR-10 and CIFAR-100 with DenseNet-DS and ResNet-DS architectures demonstrate: (1) DenseNet-DS: +8.3% faster convergence on CIFAR-10 with comparable accuracy, (2) Hybrid approach: Combining LSUV with LION-DG achieves best accuracy (81.92% on CIFAR-10), (3) ResNet-DS: Positive speedup on CIFAR-100 (+11.3%) with side-tap auxiliary design. We identify architecture-specific trade-offs and provide clear guidelines for practitioners. LION-DG is simple, requires zero hyperparameters, and adds no computational overhead.

cs.LG↗

HOLOGRAPH: Active Causal Discovery via Sheaf-Theoretic Alignment of Large Language Model Priors

Causal discovery from observational data remains fundamentally limited by identifiability constraints. Recent work has explored leveraging Large Language Models (LLMs) as sources of prior causal knowledge, but existing approaches rely on heuristic integration that lacks theoretical grounding. We introduce HOLOGRAPH, a framework that formalizes LLM-guided causal discovery through sheaf theory--representing local causal beliefs as sections of a presheaf over variable subsets. Our key insight is that coherent global causal structure corresponds to the existence of a global section, while topological obstructions manifest as non-vanishing sheaf cohomology. We propose the Algebraic Latent Projection to handle hidden confounders and Natural Gradient Descent on the belief manifold for principled optimization. Experiments on synthetic and real-world benchmarks demonstrate that HOLOGRAPH provides rigorous mathematical foundations while achieving competitive performance on causal discovery tasks with 50-100 variables. Our sheaf-theoretic analysis reveals that while Identity, Transitivity, and Gluing axioms are satisfied to numerical precision (<10^{-6}), the Locality axiom fails for larger graphs, suggesting fundamental non-local coupling in latent variable projections. Code is available at [https://github.com/hyunjun1121/holograph](https://github.com/hyunjun1121/holograph).

cs.LG↗

Defensive M2S: Training Guardrail Models on Compressed Multi-turn Conversations

Guardrail models are essential for ensuring the safety of Large Language Model (LLM) deployments, but processing full multi-turn conversation histories incurs significant computational cost. We propose Defensive M2S, a training paradigm that fine-tunes guardrail models on Multi-turn to Single-turn (M2S) compressed conversations rather than complete dialogue histories. We provide a formal complexity analysis showing that M2S reduces training cost from $O(n^2)$ to $O(n)$ for $n$-turn conversations. Empirically, on our training dataset (779 samples, avg. 10.6 turns), M2S requires only 169K tokens compared to 15.7M tokens for the multi-turn baseline -- a 93$\times$ reduction. We evaluate Defensive M2S across three guardrail model families (LlamaGuard, Nemotron, Qwen3Guard) and three compression templates (hyphenize, numberize, pythonize) on SafeDialBench, a comprehensive multi-turn jailbreak benchmark. Our best configuration, Qwen3Guard with hyphenize compression, achieves 93.8% attack detection recall while reducing inference tokens by 94.6% (from 3,231 to 173 tokens per conversation). This represents a 38.9 percentage point improvement over the baseline while dramatically reducing both training and inference costs. Our findings demonstrate that M2S compression can serve as an effective efficiency technique for guardrail deployment, enabling scalable safety screening of long multi-turn conversations.

cs.CL↗

Geometric Regularization in Mixture-of-Experts: The Disconnect Between Weights and Activations

Mixture-of-Experts (MoE) models achieve efficiency through sparse activation, but the role of geometric regularization in expert specialization remains unclear. We apply orthogonality loss to enforce expert diversity and find it fails on multiple fronts: it does not reduce weight-space overlap (MSO actually increases by up to 114%), activation-space overlap remains high (~0.6) regardless of regularization, and effects on performance are inconsistent -- marginal improvement on WikiText-103 (-0.9%), slight degradation on TinyStories (+0.9%), and highly variable results on PTB (std > 1.0). Our analysis across 7 regularization strengths reveals no significant correlation (r = -0.293, p = 0.523) between weight and activation orthogonality. These findings demonstrate that weight-space regularization neither achieves its geometric goal nor reliably improves performance, making it unsuitable for MoE diversity.

cs.LG↗