Liu Peng

Liu Peng, Yaochu Jin

A systematic, comparative investigation into the effects of low-quality data reveals a stark spectrum of robustness across modern probabilistic models. We find that autoregressive language models, from token prediction to sequence-to-sequence tasks, are remarkably resilient (for GPT-2, test NLL increases modestly from 2.87 to 3.59 despite 50% token corruption). By contrast, under the same levels of data corruption, class-conditional diffusion models degrade catastrophically (image-label consistency plummets by 56.81% relative to baseline), while classifiers show a moderate impact that diminishes with dataset scale. To explain these discrepancies, we analyze the results through a multi-perspective lens, integrating information theory, PAC learning, and gradient dynamics. These analyses suggest that robustness is heavily influenced by two key principles: the richness of conditioning information, which constrains the learning problem, and the absolute information content of the training data, which allows the signal from correct information to dominate statistical noise.

Tian Chen, Wang Ya Zhe, Liu Peng et al.

A hook is a piece of code. It checks user privacy policy before some sensitive operations happen. We propose an automated solution named Prihook for hook placement in the Android Framework. Addressing specific context-aware user privacy concerns, the hook placement in Prihook is personalized. Specifically, we design User Privacy Preference Table (UPPT) to help a user express his privacy concerns. And we leverage machine learning to discover a Potential Method Set (consisting of Sensor Data Access Methods and Sensor Control Methods) from which we can select a particular subset to put hooks. We propose a mapping from words in the UPPT lexicon to methods in the Potential Method Set. With this mapping, Prihook is able to (a) select a specific set of methods; and (b) generate and place hooks automatically. We test Prihook separately on 6 typical UPPTs representing 6 kinds of resource-sensitive UPPTs, and no user privacy violation is found. The experimental results show that the hooks placed by PriHook have small runtime overhead.