Search for: All records

Machine learning models are vulnerable to both security attacks (e.g., adversarial examples) and privacy attacks (e.g., private attribute inference). We take the first step to mitigate both the security and privacy attacks, and maintain task utility as well. Particularly, we propose an information-theoretic framework to achieve the goals through the lens of representation learning, i.e., learning representations that are robust to both adversarial examples and attribute inference adversaries. We also derive novel theoretical results under our framework, e.g., the inherent trade-off between adversarial robustness/utility and attribute privacy, and guaranteed attribute privacy leakage against attribute inference adversaries. 

Federated learning (FL) has been widely studied recently due to its property to collaboratively train data from different devices without sharing the raw data. Nevertheless, recent studies show that an adversary can still be possible to infer private information about devices' data, e.g., sensitive attributes such as income, race, and sexual orientation. To mitigate the attribute inference attacks, various existing privacy-preserving FL methods can be adopted/adapted. However, all these existing methods have key limitations: they need to know the FL task in advance, or have intolerable computational overheads or utility losses, or do not have provable privacy guarantees. We address these issues and design a task-agnostic privacy-preserving presentation learning method for FL (TAPPFL) against attribute inference attacks. TAPPFL is formulated via information theory. Specifically, TAPPFL has two mutual information goals, where one goal learns task-agnostic data representations that contain the least information about the private attribute in each device's data, and the other goal ensures the learnt data representations include as much information as possible about the device data to maintain FL utility. We also derive privacy guarantees of TAPPFL against worst-case attribute inference attacks, as well as the inherent tradeoff between utility preservation and privacy protection. Extensive results on multiple datasets and applications validate the effectiveness of TAPPFL to protect data privacy, maintain the FL utility, and be efficient as well. Experimental results also show that TAPPFL outperforms the existing defenses. 

Abstract As on land, oceans exhibit high temporal and spatial temperature variation. This “ocean weather” contributes to the physiological and ecological processes that ultimately determine the patterns of species distribution and abundance, yet is often unrecognized, especially in tropical oceans. Here, we tested the paradigm of temperature stability in shallow waters (<12.5 m) across different zones of latitude. We collated hundreds of in situ, high temporal-frequency ocean temperature time series globally to produce an intuitive measure of temperature variability, ranging in scale from quarter-diurnal to annual time spans. To estimate organismal sensitivity of ectotherms (i.e. microbes, algae, and animals whose body temperatures depend upon ocean temperature), we computed the corresponding range of biological rates (such as metabolic rate or photosynthesis) for each time span, assuming an exponential relationship. We found that subtropical regions had the broadest temperature ranges at time spans equal to or shorter than a month, while temperate and tropical systems both exhibited narrow (i.e. stable) short-term temperature range estimates. However, temperature-dependent biological rates in tropical regions displayed greater ranges than in temperate systems. Hence, our results suggest that tropical ectotherms may be relatively more sensitive to short-term thermal variability. We also highlight previously unexplained macroecological patterns that may be underpinned by short-term temperature variability.