arXiv subjects

Doyeon Kim

Publications and source records attributed to Doyeon Kim.

At least 19 recordsLinked to original sources

arXiv2026

When Feedback Becomes a Backdoor: Tracing Backdoor Skill Formation in Text-Space Skill Optimization

Reference answers are a supervision channel in text-space Skill optimization, which produces reusable text artifacts called Skills that guide future agent behavior. Poisoning those answers can induce a persistent trigger-conditioned rule without giving the attacker access to the model, optimizer, or Skill file. We study this reference-feedback poisoning in SkillOpt by tracing an attacker-supplied trigger--payload mapping from reflection to held-out activation. In SearchQA, where the payload replaces a short answer, the natural attack retains an exact rule in all three seeds; each final Skill emits the payload on 200/200 triggered questions and 0/200 clean questions. In seed 42, removing the rule eliminates activation, and an in-run comparison shows similar clean accuracy to benign SkillOpt (169/200 versus 171/200). In SpreadsheetBench, where the payload requires an additional workbook edit, no natural run retains an exact rule across three seeds. Diagnostic interventions locate two successive bottlenecks: the Analyst abstracts away required literals, and, once literal preservation is enabled, edit-budget ranking can remove exact edits before candidate construction. Poisoned references can therefore create persistent backdoors, but exposure alone does not ensure that the induced rule is formed, selected, and retained.

cs.CR↗

arXiv2026

HERALD: High-Throughput Block Diffusion LLM Inference via CPU-GPU Cooperative KV Cache Retrieval

Block diffusion LLMs (block dLLMs) improve decode throughput by generating tokens in parallel, and are emerging as a next-generation alternative to autoregressive LLMs. At long contexts, however, the KV cache remains a major throughput bottleneck in both memory bandwidth and capacity. For autoregressive LLMs, dynamic sparse attention is often combined with KV offloading to alleviate both bottlenecks, keeping the full KV cache in CPU DRAM and fetching only a critical subset to the GPU. However, KV subset selection and retrieval over bandwidth-limited PCIe incur substantial offloading overheads that often stall the GPU. We present HERALD, the first dynamic sparse attention method for block dLLMs that supports KV offloading. HERALD exploits a property of block dLLMs: a critical KV subset selected once per block is reused across all of its denoising steps, so the next block's selection can be hidden behind the current block's denoising. HERALD realizes this with a two-stage selection algorithm co-designed with its execution system. While the GPU denoises the current block, the CPU speculatively searches the full KV cache for the next block with a single query merged from the current block's B queries, cutting the selection compute by Bx; the GPU then refines the candidates into the final top-K subset with the next block's real queries. At K=1024, HERALD achieves near-lossless accuracy relative to Dense on LongBench and reasoning tasks across four block dLLMs, and improves decode throughput by up to 3.70x over Dense GPU-only inference, 2.59x over MAGE, a state-of-the-art block-dLLM sparse attention method, and 8.38x over MAGE-Offload, its direct KV-offloading extension.

cs.LG↗

arXiv2026

A^2Agent: Action-Aware Reinforcement Learning for Repository-Level Code Localization Agents

Localizing issue-relevant code regions is a critical step in automated software engineering. However, due to their reliance on sparse trajectory-level signals, existing methods cannot identify which per-turn actions are effective and often discover correct code regions during exploration but fail to commit them. To address these limitations, we propose an action-aware reinforcement learning method that combines a per-turn reward sequence rewarding both the discovery and commitment of gold code regions with an action-level advantage estimation scheme that isolates each action's credit by grouping turns sharing the same exploration context. Extensive evaluations show that our method improves the average F1 over the state-of-the-art (SOTA) by 1.58% on SWE-Bench Verified and 8.55% on SWE-Bench Pro, with our 4B model outperforming baselines up to 8x larger. Our code is available at https://github.com/donian00/A2Agent.

cs.CL↗

arXiv2026

Trust Boundary Semantic Gaps: A Multi-dimensional Analysis and Mitigation for Security-by-Design

Modern systems use format-, protocol-, and signature-based mechanisms before accepting artifacts across trust boundaries. These mechanisms are necessary: they show that an artifact is well formed, protocol-compliant, or properly authenticated. They do not, however, show that the artifact satisfies the semantic security properties required by the receiving domain. A signed update or an authenticated token may therefore be accepted yet enable compromise. We call this condition a Trust Boundary Semantic Gap (TBSG): an artifact crosses a trust boundary and passes correctly implemented syntactic validation, but the assertions established by that pass are insufficient to satisfy the receiving domain's security requirements. TBSG concerns what remains unestablished after a syntactic pass, not absent checks or implementation bugs. Analyzing 75 publicly reported security incidents (2014-2025) at the boundary level, we organize semantic misalignment into a four-dimensional analysis model: Identity, Spatial, Temporal, and Interpretation (MDTBSG). Building on it, we develop Trust Boundary Semantic Analysis and Mitigation (TBSAM), a design-time framework that identifies TBSGs from design specifications, prioritizes them, traces propagated gaps to their originating boundary, and maps each to candidate architectural controls. We apply TBSAM to a retrospective reconstruction of the SolarWinds/SUNBURST supply-chain attack, showing how it makes receiving-domain assumptions explicit, separates locally originating from propagated gaps, and identifies controls that interrupt the path. These results suggest that syntactic validation, while necessary, is not sufficient at trust boundaries, and that making trust-boundary assumptions explicit can complement Security-by-Design.

cs.CR↗

arXiv2026

MAGE: All-[MASK] Block Already Knows Where to Look in Block Diffusion LLM

Block diffusion LLMs are an emerging paradigm for parallel language generation, but their KV caching makes memory access the dominant bottleneck in long-context inference. Sparse attention, which attends only to a small KV subset per query, can reduce this latency with minimal accuracy loss. In block diffusion, however, the B tokens of each block must share a single KV subset, and we show this per-block constraint degrades existing sparse KV estimators by up to 25% in recall. We address this challenge by exploiting a property that emerges from the block-diffusion training objective: it aligns the block-average query across denoising steps, so the All-[MASK] block at the first step already reveals the per-block KV subset for the entire trajectory. We exploit this in MAGE ([MASK]-Guided Sparse Attention), a training-free method that runs one exact attention pass at the first step and reuses its top-k index sets for all remaining steps within the block. Across three block-diffusion families on LongBench, MAGE matches Exact Attention at k=512 with near-lossless accuracy, achieves up to 6.82x end-to-end speedup at 128K context, and runs up to 3.35x and 2.28x faster than Quest and SparseD, designed for AR LLMs and fully bidirectional diffusion LLMs, respectively.

cs.LG↗

arXiv2026

IntentCUA: Learning Intent-level Representations for Skill Abstraction and Multi-Agent Planning in Computer-Use Agents

Computer-use agents operate over long horizons under noisy perception, multi-window contexts, evolving environment states. Existing approaches, from RL-based planners to trajectory retrieval, often drift from user intent and repeatedly solve routine subproblems, leading to error accumulation and inefficiency. We present IntentCUA, a multi-agent computer-use framework designed to stabilize long-horizon execution through intent-aligned plan memory. A Planner, Plan-Optimizer, and Critic coordinate over shared memory that abstracts raw interaction traces into multi-view intent representations and reusable skills. At runtime, intent prototypes retrieve subgroup-aligned skills and inject them into partial plans, reducing redundant re-planning and mitigating error propagation across desktop applications. In end-to-end evaluations, IntentCUA achieved a 74.83% task success rate with a Step Efficiency Ratio of 0.91, outperforming RL-based and trajectory-centric baselines. Ablations show that multi-view intent abstraction and shared plan memory jointly improve execution stability, with the cooperative multi-agent loop providing the largest gains on long-horizon tasks. These results highlight that system-level intent abstraction and memory-grounded coordination are key to reliable and efficient desktop automation in large, dynamic environments.

cs.AI↗

arXiv2025

MF-PAM: Accurate Pitch Estimation through Periodicity Analysis and Multi-level Feature Fusion

We introduce Multi-level feature Fusion-based Periodicity Analysis Model (MF-PAM), a novel deep learning-based pitch estimation model that accurately estimates pitch trajectory in noisy and reverberant acoustic environments. Our model leverages the periodic characteristics of audio signals and involves two key steps: extracting pitch periodicity using periodic non-periodic convolution (PNP-Conv) blocks and estimating pitch by aggregating multi-level features using a modified bi-directional feature pyramid network (BiFPN). We evaluate our model on speech and music datasets and achieve superior pitch estimation performance compared to state-of-the-art baselines while using fewer model parameters. Our model achieves 99.20 % accuracy in pitch estimation on a clean musical dataset. Overall, our proposed model provides a promising solution for accurate pitch estimation in challenging acoustic environments and has potential applications in audio signal processing.

eess.AS↗

arXiv2025

Doppelganger Method: Breaking Role Consistency in LLM Agent via Prompt-based Transferable Adversarial Attack

Since the advent of large language models, prompt engineering now enables the rapid, low-effort creation of diverse autonomous agents that are already in widespread use. Yet this convenience raises urgent concerns about the safety, robustness, and behavioral consistency of the underlying prompts, along with the pressing challenge of preventing those prompts from being exposed to user's attempts. In this paper, we propose the ''Doppelganger method'' to demonstrate the risk of an agent being hijacked, thereby exposing system instructions and internal information. Next, we define the ''Prompt Alignment Collapse under Adversarial Transfer (PACAT)'' level to evaluate the vulnerability to this adversarial transfer attack. We also propose a ''Caution for Adversarial Transfer (CAT)'' prompt to counter the Doppelganger method. The experimental results demonstrate that the Doppelganger method can compromise the agent's consistency and expose its internal information. In contrast, CAT prompts enable effective defense against this adversarial attack.

cs.AI↗

arXiv2023

A Worker-Task Specialization Model for Crowdsourcing: Efficient Inference and Fundamental Limits

Crowdsourcing system has emerged as an effective platform for labeling data with relatively low cost by using non-expert workers. Inferring correct labels from multiple noisy answers on data, however, has been a challenging problem, since the quality of the answers varies widely across tasks and workers. Many existing works have assumed that there is a fixed ordering of workers in terms of their skill levels, and focused on estimating worker skills to aggregate the answers from workers with different weights. In practice, however, the worker skill changes widely across tasks, especially when the tasks are heterogeneous. In this paper, we consider a new model, called $d$-type specialization model, in which each task and worker has its own (unknown) type and the reliability of each worker can vary in the type of a given task and that of a worker. We allow that the number $d$ of types can scale in the number of tasks. In this model, we characterize the optimal sample complexity to correctly infer the labels within any given accuracy, and propose label inference algorithms achieving the order-wise optimal limit even when the types of tasks or those of workers are unknown. We conduct experiments both on synthetic and real datasets, and show that our algorithm outperforms the existing algorithms developed based on more strict model assumptions.

cs.HC↗

arXiv2023

HD-DEMUCS: General Speech Restoration with Heterogeneous Decoders

This paper introduces an end-to-end neural speech restoration model, HD-DEMUCS, demonstrating efficacy across multiple distortion environments. Unlike conventional approaches that employ cascading frameworks to remove undesirable noise first and then restore missing signal components, our model performs these tasks in parallel using two heterogeneous decoder networks. Based on the U-Net style encoder-decoder framework, we attach an additional decoder so that each decoder network performs noise suppression or restoration separately. We carefully design each decoder architecture to operate appropriately depending on its objectives. Additionally, we improve performance by leveraging a learnable weighting factor, aggregating the two decoder output waveforms. Experimental results with objective metrics across various environments clearly demonstrate the effectiveness of our approach over a single decoder or multi-stage systems for general speech restoration task.

eess.AS↗

arXiv2023

Context-Preserving Two-Stage Video Domain Translation for Portrait Stylization

Portrait stylization, which translates a real human face image into an artistically stylized image, has attracted considerable interest and many prior works have shown impressive quality in recent years. However, despite their remarkable performances in the image-level translation tasks, prior methods show unsatisfactory results when they are applied to the video domain. To address the issue, we propose a novel two-stage video translation framework with an objective function which enforces a model to generate a temporally coherent stylized video while preserving context in the source video. Furthermore, our model runs in real-time with the latency of 0.011 seconds per frame and requires only 5.6M parameters, and thus is widely applicable to practical real-world applications.

cs.CV↗

arXiv2023

Fix the Noise: Disentangling Source Feature for Transfer Learning of StyleGAN

Transfer learning of StyleGAN has recently shown great potential to solve diverse tasks, especially in domain translation. Previous methods utilized a source model by swapping or freezing weights during transfer learning, however, they have limitations on visual quality and controlling source features. In other words, they require additional models that are computationally demanding and have restricted control steps that prevent a smooth transition. In this paper, we propose a new approach to overcome these limitations. Instead of swapping or freezing, we introduce a simple feature matching loss to improve generation quality. In addition, to control the degree of source features, we train a target model with the proposed strategy, FixNoise, to preserve the source features only in a disentangled subspace of a target feature space. Owing to the disentangled feature space, our method can smoothly control the degree of the source features in a single model. Extensive experiments demonstrate that the proposed method can generate more consistent and realistic images than previous works.

cs.CV↗

arXiv2023

Fix the Noise: Disentangling Source Feature for Controllable Domain Translation

Recent studies show strong generative performance in domain translation especially by using transfer learning techniques on the unconditional generator. However, the control between different domain features using a single model is still challenging. Existing methods often require additional models, which is computationally demanding and leads to unsatisfactory visual quality. In addition, they have restricted control steps, which prevents a smooth transition. In this paper, we propose a new approach for high-quality domain translation with better controllability. The key idea is to preserve source features within a disentangled subspace of a target feature space. This allows our method to smoothly control the degree to which it preserves source features while generating images from an entirely new domain using only a single model. Our extensive experiments show that the proposed method can produce more consistent and realistic images than previous works and maintain precise controllability over different levels of transformation. The code is available at https://github.com/LeeDongYeun/FixNoise.

cs.CV↗

arXiv2023

Feature Structure Distillation with Centered Kernel Alignment in BERT Transferring

Knowledge distillation is an approach to transfer information on representations from a teacher to a student by reducing their difference. A challenge of this approach is to reduce the flexibility of the student's representations inducing inaccurate learning of the teacher's knowledge. To resolve it in transferring, we investigate distillation of structures of representations specified to three types: intra-feature, local inter-feature, global inter-feature structures. To transfer them, we introduce feature structure distillation methods based on the Centered Kernel Alignment, which assigns a consistent value to similar features structures and reveals more informative relations. In particular, a memory-augmented transfer method with clustering is implemented for the global structures. The methods are empirically analyzed on the nine tasks for language understanding of the GLUE dataset with Bidirectional Encoder Representations from Transformers (BERT), which is a representative neural language model. In the results, the proposed methods effectively transfer the three types of structures and improve performance compared to state-of-the-art distillation methods. Indeed, the code for the methods is available in https://github.com/maroo-sky/FSD.

cs.CL↗

arXiv2022

Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth

Depth estimation from a single image is an important task that can be applied to various fields in computer vision, and has grown rapidly with the development of convolutional neural networks. In this paper, we propose a novel structure and training strategy for monocular depth estimation to further improve the prediction accuracy of the network. We deploy a hierarchical transformer encoder to capture and convey the global context, and design a lightweight yet powerful decoder to generate an estimated depth map while considering local connectivity. By constructing connected paths between multi-scale local features and the global decoding stream with our proposed selective feature fusion module, the network can integrate both representations and recover fine details. In addition, the proposed decoder shows better performance than the previously proposed decoders, with considerably less computational complexity. Furthermore, we improve the depth-specific augmentation method by utilizing an important observation in depth estimation to enhance the model. Our network achieves state-of-the-art performance over the challenging depth dataset NYU Depth V2. Extensive experiments have been conducted to validate and show the effectiveness of the proposed approach. Finally, our model shows better generalisation ability and robustness than other comparative models.

cs.CV↗

arXiv2022

Learning Audio-Text Agreement for Open-vocabulary Keyword Spotting

In this paper, we propose a novel end-to-end user-defined keyword spotting method that utilizes linguistically corresponding patterns between speech and text sequences. Unlike previous approaches requiring speech keyword enrollment, our method compares input queries with an enrolled text keyword sequence. To place the audio and text representations within a common latent space, we adopt an attention-based cross-modal matching approach that is trained in an end-to-end manner with monotonic matching loss and keyword classification loss. We also utilize a de-noising loss for the acoustic embedding network to improve robustness in noisy environments. Additionally, we introduce the LibriPhrase dataset, a new short-phrase dataset based on LibriSpeech for efficiently training keyword spotting models. Our proposed method achieves competitive results on various evaluation sets compared to other single-modal and cross-modal baselines.

eess.AS↗

arXiv2022

Tectonics of Cerberus Fossae unveiled by marsquakes

The InSight mission has measured Mars' seismicity since February 2018 and has allowed to investigate tectonics on another planet. Seismic data shows that most of the widely distributed surface faults are not seismically active, and that seismicity is mostly originating from a single graben structure, the Cerberus Fossae. We show that both major families of marsquakes characterized by low and high frequency content, LF and HF events respectively, are located on central and eastern parts of this graben system. LF hypocenters are located at 15-50 km depth and the spectral character suggests a structurally weak, potentially warm source region consistent with recent volcanic activity at those depths. HF marsquakes occur in the brittle, shallow part of the crust and might originate in fault planes associated with the graben flanks. Estimated magnitudes are between 2.8 and 3.8, resulting in a total seismic moment release within Cerberus Fossae of 1.4-5.6 $\times10^{15}$ Nm/yr, or at least half of the observed value of the entire planet. Our findings confirm that Cerberus Fossae represents a unique tectonic setting shaped by current day volcanic processes, with implications for minimum local heat flow.

astro-ph.EP↗

arXiv2022

Self-supervised GAN Detector

Although the recent advancement in generative models brings diverse advantages to society, it can also be abused with malicious purposes, such as fraud, defamation, and fake news. To prevent such cases, vigorous research is conducted to distinguish the generated images from the real images, but challenges still remain to distinguish the unseen generated images outside of the training settings. Such limitations occur due to data dependency arising from the model's overfitting issue to the training data generated by specific GANs. To overcome this issue, we adopt a self-supervised scheme to propose a novel framework. Our proposed method is composed of the artificial fingerprint generator reconstructing the high-quality artificial fingerprints of GAN images for detailed analysis, and the GAN detector distinguishing GAN images by learning the reconstructed artificial fingerprints. To improve the generalization of the artificial fingerprint generator, we build multiple autoencoders with different numbers of upconvolution layers. With numerous ablation studies, the robust generalization of our method is validated by outperforming the generalization of the previous state-of-the-art algorithms, even without utilizing the GAN images of the training dataset.

cs.CV↗