Institution profile

University of Seoul

Academic institutionasia · kr
Official website
Research library87linked papers
Opportunities0open roles
Selected work

Representative Papers

Zero2Text: Zero-Training Cross-Domain Inversion Attacks on Textual Embeddings

Feb 02, 2026

This work addresses the challenge of embedding inversion under strict black-box and cross-domain settings, where existing methods struggle to preserve vector database privacy due to their reliance on extensive queries or in-domain training data. We propose the first training-free cross-domain embedding inversion framework, which leverages a recursive online alignment mechanism that integrates large language model priors with dynamic ridge regression to generate text aligned with target embeddings in real time—without any training or prior knowledge of the target domain. Our approach eliminates dependence on static datasets or high query budgets and exposes limitations of conventional defenses such as differential privacy. Experiments demonstrate significant improvements over baselines across multiple benchmarks including MS MARCO, achieving a 1.8× increase in ROUGE-L and a 6.4× gain in BLEU-2 against OpenAI models, successfully reconstructing original sentences from unseen domains.

1 citationsRead paper

Robust Adaptation of Foundation Models with Black-Box Visual Prompting

Jul 04, 2024arXiv.org

Black-box large models (e.g., API-hosted pre-trained models) are inaccessible in terms of parameters and architecture, and suffer from severe GPU memory constraints. Method: We propose BlackVIP, a vision-language prompting method requiring no internal model information. It employs an input-dependent prompt generation mechanism coupled with SPSA-GC-based gradient estimation, enabling memory-efficient, backpropagation-free adaptation. We further introduce BlackVIP-SE—a lightweight variant—and establish, for the first time, a theoretical connection between visual prompting and randomized smoothing robustness, formally explaining its improved adversarial robustness. Contribution/Results: Evaluated across 19 cross-domain datasets, BlackVIP significantly reduces GPU memory consumption and computational overhead while enhancing out-of-distribution generalization and adversarial robustness—without accessing model internals or gradients.

1 citationsRead paper

FFNet: MetaMixer-based Efficient Convolutional Mixer Design

Jun 04, 2024

Vision backbone networks suffer from redundancy in feed-forward network (FFN) design and high computational overhead in attention mechanisms. Method: We propose FFNet—a purely convolutional architecture grounded in the novel “FFNification” paradigm—unifying the QKV abstraction framework with FFN structure. It replaces QKV linear projections and Softmax with large-kernel depthwise convolutions, substitutes non-linear activation with GELU, and integrates ConvNeXt-style residual blocks to construct a lightweight, efficient token mixer. We further introduce MetaMixer, a generic mixer architecture that decouples token-mixing from channel-mixing. Contribution/Results: FFNet establishes the foundational insight that the QKV framework is more fundamental than its specific instantiations (e.g., self-attention or FFN). It outperforms same-scale specialized models across image classification, detection, and segmentation, while significantly improving inference speed and parameter efficiency—demonstrating the general applicability and effectiveness of the QKV framework within pure convolutional architectures.

1 citationsRead paper
Recent publications

Latest Papers

MIFA: An MILP-based Framework for Improving Differential Fault Attacks

Aug 07, 2026

This work addresses the inefficiency of differential fault analysis (DFA) by proposing the first DFA framework based on mixed-integer linear programming (MILP). For the first time, MILP is employed to systematically search for differential trails with a unique solution, combined with bit-level single-bit flip fault modeling to optimize both the location and number of injected faults. The approach enables attacks on deeper-round implementations and allows theoretical computation of the minimal number of faults required to recover the secret key. When applied to the DEFAULT block cipher, the method uniquely recovers the full key with only three faults in the sixth-to-last round and two faults each in the seventh- and eighth-to-last rounds, significantly outperforming existing DFA results and effectively breaking the cipher’s claimed DFA resistance.

0 citationsRead paper

Attention from Above: A Multimodal Model for Drone-Based Object Localization

Jul 20, 2026

This work addresses the limited localization accuracy of small objects in drone-based scenarios under text-prompt guidance by proposing an enhanced approach built upon the YOLO-World framework. The method introduces a novel A2C2f module to replace the original C2f layer and integrates an attention mechanism into the backbone network to strengthen local feature representation, thereby significantly improving multimodal detection performance for small and boundary-sharp objects. Employing a parallel multimodal fusion architecture, the proposed model achieves an mAP@0.5 of 35.2% and an F1 score of 39.4% on the VisDrone dataset, substantially outperforming the baseline YOLO-World model.

0 citationsRead paper