About
Research & Development
Prompt-Only OverThinking: A Stealthy Approach to Manipulate LLM Reasoning Efficiency
TLDR: The paper introduces POT (Prompt-Only OverThinking), a novel black-box attack framework that uses LLM-based iterative optimization to create subtle, natural adversarial prompts. These...
Research & Development
Enhancing Data Privacy with LLMs: A New Framework for Synthetic Text Rewriting
TLDR: This research introduces the STYLE-MST Rewriter, a reinforcement learning framework that fine-tunes large language models (LLMs) to generate privacy-preserving synthetic text. It addresses...
Research & Development
Securing Autonomous Vehicle Perception: A Lightweight Defense Against LiDAR Attacks
TLDR: A new research paper introduces DU-AP, an efficient and lightweight defense mechanism against adversarial attacks on LiDAR segmentation systems used in autonomous vehicles....
Research & Development
Bridging Vision and Language: The Evolution of Object Detection with Advanced AI Models
TLDR: This research paper provides an in-depth review of how Multimodal Large Vision-Language Models (LVLMs) are transforming object detection. It details their architectural innovations,...
Research & Development
Demographic Biases in AI Face Recognition: An Empirical Look at LVLMs
TLDR: The DemoBias study empirically evaluates demographic biases in Large Vision Language Models (LVLMs) for biometric face recognition. Researchers fine-tuned LLaVA, BLIP-2, and PaliGemma...
Research & Development
Advanced Imaging for Live Cells: CellINR Tackles Photo-induced Artifacts in 4D Fluorescence Microscopy
TLDR: CellINR is a novel framework that utilizes implicit neural representations, blind convolution, and structure amplification to effectively remove photo-induced artifacts in 4D live...