6 papers
Prompt Attack Detection with LLM-as-a-Judge and Mixture-of-Models
Hieu Xuan Le, Benjamin Goh, Quy Anh Tang
Prompt attacks, including jailbreaks and prompt injections, pose a critical security risk to Large Language Model (LLM) systems. In production, guardrails must mitigate these attac…
LoRIF: Low-Rank Influence Functions for Scalable Training Data Attribution
Shuangqi Li, Hieu Le, Jingyi Xu +1
Training data attribution (TDA) identifies which training examples most influenced a model's prediction. Influence function methods are a theoretically grounded family of TDA metho…
Learning to Weight Parameters for Training Data Attribution
Shuangqi Li, Hieu Le, Jingyi Xu +1
We study gradient-based data attribution, aiming to identify which training examples most influence a given output. Existing methods for this task either treat network parameters u…
All Seeds Are Not Equal: Enhancing Compositional Text-to-Image Generation with Reliable Random Seeds
Shuangqi Li, Hieu Le, Jingyi Xu +1
Text-to-image diffusion models have demonstrated remarkable capability in generating realistic images from arbitrary text prompts. However, they often produce inconsistent results…
QT-DoG: Quantization-aware Training for Domain Generalization
Saqib Javed, Hieu Le, Mathieu Salzmann
A key challenge in Domain Generalization (DG) is preventing overfitting to source domains, which can be mitigated by finding flatter minima in the loss landscape. In this work, we…
Controlling the Fidelity and Diversity of Deep Generative Models via Pseudo Density
Shuangqi Li, Chen Liu, Tong Zhang +3
We introduce an approach to bias deep generative models, such as GANs and diffusion models, towards generating data with either enhanced fidelity or increased diversity. Our approa…