Institution profile

University of Texas at Dallas

Academic institutionnorthamerica · us
Official website
Research library546linked papers
Opportunities0open roles
Selected work

Representative Papers

Jailbreaking and Mitigation of Vulnerabilities in Large Language Models

Oct 20, 2024arXiv.org

Existing research on large language models (LLMs) lacks a unified taxonomy for prompt injection and jailbreaking attacks, and insufficiently evaluates defenses under dynamic, interactive scenarios. Method: We propose the first four-dimensional attack taxonomy—spanning prompt-level, model-level, multimodal, and multilingual pathways—and develop a robust alignment framework tailored to interactive settings, alongside a novel automated jailbreaking detection method. We further conduct systematic defense benchmarking, bias diagnosis of existing evaluation benchmarks, and multi-dimensional security measurement. Contribution/Results: Our analysis reveals critical failure modes of current defenses in dynamic interactions, identifies key research gaps—including ethical implications and data bias—and delivers the first comprehensive technical roadmap for LLM safety alignment.

7 citationsRead paper

TP‑Blend: Textual‑Prompt Attention Pairing for Precise Object‑Style Blending in Diffusion Models

Jan 12, 2026Trans. Mach. Learn. Res.

This work proposes TP-Blend, a lightweight, training-free framework that addresses the longstanding challenge in text-guided diffusion-based image editing: the simultaneous and precise injection of novel objects and styles. TP-Blend achieves decoupled control by synergistically combining Cross-Attention Object Fusion (CAOF) and Self-Attention Style Fusion (SASF) within a single denoising step, enabling concurrent manipulation of content and texture. The method further enhances fidelity through entropy-regularized optimal transport, detail-sensitive instance normalization, and high–low frequency decomposition, effectively preserving multi-head feature correlations while boosting fine-grained detail retention. Extensive experiments demonstrate that TP-Blend significantly outperforms existing approaches in high-resolution image generation, achieving state-of-the-art performance in content fidelity, perceptual quality, and inference speed.

4 citationsRead paper

HiDVFS: A Hierarchical Multi-Agent DVFS Scheduler for OpenMP DAG Workloads

Jan 10, 2026arXiv.org

Existing DVFS scheduling approaches struggle to achieve fine-grained, coordinated optimization of performance, energy consumption, and thermal behavior for OpenMP DAG tasks during parallel execution. This work proposes HiDVFS, a hierarchical multi-agent reinforcement learning scheduler that introduces, for the first time, a hierarchical multi-agent architecture to DVFS. HiDVFS employs three cooperative agents to jointly manage task-core-frequency assignment, temperature-aware core-set selection, and task prioritization under resource contention. By integrating task profiling data, real-time thermal feedback, and a makespan-oriented reward function augmented with energy-efficiency and thermal regularization terms, HiDVFS achieves, on the NVIDIA Jetson TX2 platform using the BOTS benchmark suite, an average speedup of 3.95× and 47.1% energy reduction compared to GearDVFS, with peak improvements reaching 3.44× speedup and 50.4% energy savings.

3 citationsRead paper

Deblur4DGS: 4D Gaussian Splatting from Blurry Monocular Video

Dec 09, 2024arXiv.org

Existing 4D reconstruction methods for blurry monocular videos struggle to accurately model continuous dynamics during exposure due to coupled camera jitter and object motion, leading to motion blur. This work introduces the first 4D extension of 3D Gaussian Splatting—termed 4D Gaussian Splatting—designed specifically to address motion blur in dynamic scenes. Instead of conventional continuous motion representations, it models scene elements via explicit trajectory parameterization; replaces explicit motion estimation with exposure-time-aware modeling; employs blur-aware, variable-canonical Gaussians; and incorporates exposure regularization alongside multi-frame and multi-resolution consistency constraints. Evaluated on four downstream tasks—novel view synthesis, video deblurring, frame interpolation, and video stabilization—the framework consistently outperforms state-of-the-art 4D reconstruction approaches, enabling high-fidelity dynamic scene reconstruction and effective enhancement of multi-view blurry video sequences.

2 citations1 influentialRead paper

ZeroDVFS: Zero-Shot LLM-Guided Core and Frequency Allocation for Embedded Platforms

Jan 13, 2026

This work addresses the limitations of traditional dynamic voltage and frequency scaling (DVFS) and task-core allocation methods, which rely on heuristics or offline profiling, struggle to generalize to unseen workloads, and neglect stall time—leading to suboptimal energy efficiency and thermal management. To overcome these challenges, the paper proposes the first large language model (LLM)-guided, zero-shot multi-agent reinforcement learning framework for runtime scheduling. The approach leverages an LLM to extract 13-dimensional code-level semantic features from OpenMP programs and integrates hierarchical multi-agent action decomposition, regression-based environment modeling, and a Dyna-Q architecture to enable workload-agnostic scheduling without prior profiling. Experiments on Jetson TX2/Orin NX, RubikPi, and Intel Core i7 platforms demonstrate a 7.09× improvement in energy efficiency and a 4× reduction in task completion time compared to the Linux ondemand scheduler, with the first scheduling decision made 8,300× faster than conventional tabular methods.

2 citationsRead paper
Recent publications

Latest Papers