Search for: All records

Label differential privacy is a relaxation of differential privacy for machine learning scenarios where the labels are the only sensitive information that needs to be protected in the training data. For example, imagine a survey from a participant in a university class about their vaccination status. Some attributes of the students are publicly available but their vaccination status is sensitive information and must remain private. Now if we want to train a model that predicts whether a student has received vaccination using only their public information, we can use label-DP. Recent works on label-DP use different ways of adding noise to the labels in order to obtain label-DP models. In this work, we present novel techniques for training models with label-DP guarantees by leveraging unsupervised learning and semi-supervised learning, enabling us to inject less noise while obtaining the same privacy, therefore achieving a better utility-privacy trade-off. We first introduce a framework that starts with an unsupervised classifier f0 and dataset D with noisy label set Y , reduces the noise in Y using f0 , and then trains a new model f using the less noisy dataset. Our noise reduction strategy uses the model f0 to remove the noisy labels that are incorrect with high probability. Then we use semi-supervised learning to train a model using the remaining labels. We instantiate this framework with multiple ways of obtaining the noisy labels and also the base classifier. As an alternative way to reduce the noise, we explore the effect of using unsupervised learning: we only add noise to a majority voting step for associating the learned clusters with a cluster label (as opposed to adding noise to individual labels); the reduced sensitivity enables us to add less noise. Our experiments show that these techniques can significantly outperform the prior works on label-DP. 

Abstract Rhodopsin photosystems convert light energy into electrochemical gradients used by the cell to produce ATP, or for other energy-demanding processes. While these photosystems are widespread in the ocean and have been identified in diverse microbial taxonomic groups, their physiological role in vivo has only been studied in few marine bacterial strains. Recent metagenomic studies revealed the presence of rhodopsin genes in the understudied Verrucomicrobiota phylum, yet their distribution within different Verrucomicrobiota lineages, their diversity, and function remain unknown. In this study, we show that more than 7% of Verrucomicrobiota genomes (n = 2916) harbor rhodopsins of different types. Furthermore, we describe the first two cultivated rhodopsin-containing strains, one harboring a proteorhodopsin gene and the other a xanthorhodopsin gene, allowing us to characterize their physiology under laboratory-controlled conditions. The strains were isolated in a previous study from the Eastern Mediterranean Sea and read mapping of 16S rRNA gene amplicons showed the highest abundances of these strains at the deep chlorophyll maximum (source of their inoculum) in winter and spring, with a substantial decrease in summer. Genomic analysis of the isolates suggests that motility and degradation of organic material, both energy demanding functions, may be supported by rhodopsin phototrophy in Verrucomicrobiota. Under culture conditions, we show that rhodopsin phototrophy occurs under carbon starvation, with light-mediated energy generation supporting sugar transport into the cells. Overall, this study suggests that photoheterotrophic Verrucomicrobiota may occupy an ecological niche where energy harvested from light enables bacterial motility toward organic matter and supports nutrient uptake.