While significant progress has been made separately on analytics systems for scalable stochastic gradient descent (SGD) and private SGD, none of the major scalable analytics frameworks have incorporated differentially private SGD. There are two inter-related issues for this disconnect between research and practice: (1) low model accuracy due to added noise to guarantee privacy, and (2) high development and runtime overhead of the private algorithms. This paper takes a first step to remedy this disconnect and proposes a private SGD algorithm to address both issues in an integrated manner. In contrast to the whitebox approach adopted by previous work, we revisit and use the classical technique of output perturbation to devise a novel "bolt-on" approach to private SGD. While our approach trivially addresses (2), it makes (1) even more challenging. We address this challenge by providing a novel analysis of the L 2 -sensitivity of SGD, which allows, under the same privacy guarantees, better convergence of SGD when only a constant number of passes can be made over the data. We integrate our algorithm, as well as other state-of-the-art differentially private SGD, into Bismarck, a popular scalable SGD-based analytics system on top of an RDBMS. Extensive experiments show that our algorithm can be easily integrated, incurs virtually no overhead, scales well, and most importantly, yields substantially better (up to 4X) test accuracy than the state-of-the-art algorithms on many real datasets.
To explore the effect of added graphene sheets (GNs) and added perylene-3,4,9,10-tetracarboxylic dianhydride (PTCDA) on the tribological and mechanical performances of polyimide (PI) matrix, GNs and PTCDA reinforced PI-based composites were synthesized via the blending method. The tribological properties of GNs/PTCDA/PI (GAPI) composites with different weight ratios of GNs and PTCDA under dry sliding, deionized water lubrication, and kerosene lubrication were comparatively investigated. A synergism was found between GNs and PTCDA; this synergism endowed filled PI composites with a lower friction coefficient and showed an improved wear rate under different lubrication conditions, especially when the weight ratio of GNs and PTCDA was 1:1 (GAPI-1). Under dry sliding, deionized water lubrication, and kerosene lubrication, the friction coefficient of GAPI-1 composites decreased by 41.1%, 70%, and 35.7%, respectively, while the wear rate decreased by 39%, 50%, and 25.1%, respectively. Meanwhile, the tensile strength, tensile modulus, and the elongation at break of GAPI-1 films increased by 40.8%, 51.3%, and 49.2%, respectively, relative to those of pure PI. We anticipate that this work can be used to exploit a simple and effective method for preparing materials for bearings and transmission parts that possess good tribological properties under harsh lubrication conditions.
Data compression is a popular technique for improving the efficiency of data processing workloads such as SQL queries and more recently, machine learning (ML) with classical batch gradient methods. But the efficacy of such ideas for mini-batch stochastic gradient descent (MGD), arguably the workhorse algorithm of modern ML, is an open question. MGD's unique data access pattern renders prior art, including those designed for batch gradient methods, less effective. We fill this crucial research gap by proposing a new lossless compression scheme we call tuple-oriented compression (TOC) that is inspired by an unlikely source, the string/text compression scheme Lempel-Ziv-Welch, but tailored to MGD in a way that preserves tuple boundaries within mini-batches. We then present a suite of novel compressed matrix operation execution techniques tailored to the TOC compression scheme that operate directly over the compressed data representation and avoid decompression overheads. An extensive empirical evaluation with real-world datasets shows that TOC consistently achieves substantial compression ratios by up to 51x and reduces runtimes for MGD workloads by up to 10.2x in popular ML systems. * These authors are currently at Google
Stress can affect people’s judgment and make them take risky decisions. Abnormal decision-making behavior is a core symptom of psychiatric disorders, such as anxiety, depression, and substance abuse. However, the neuronal mechanisms underlying such impairments are largely unknown. The anterior insular cortex (AIC) is a crucial structure to integrate sensory information with emotional and motivational states. These properties suggest that AIC can influence a subjective prediction in decision-making. In this study, we demonstrated that stressed mice prefer to take more risky choices than control mice using a gambling test. Manipulating the neural activity of AIC or selectively inhibiting the AIC-BLA pathway with chemogenetic intervention resulted in alterations in risk decision-making in mice. Different sexes may have different decision-making strategies in risky situations. Endogenous estrogen levels affect emotional cognition by modulating the stress system function in women. We observed decision-making behavior in mice of different sexes with or without stress experience. The result showed that female mice did not change their choice strategy with increasing risk/reward probability and performed a lower risk preference than male mice after stress. Using the pharmacological method, we bilaterally injected an estrogen receptor (ER) antagonist that resulted in more risky behavior and decreased synaptic plasticity in the AIC of female mice. Our study suggested that the AIC is a crucial region involved in stress-induced alteration of decision-making, and estrogen in the AIC may regulate decision-making behavior by regulating synaptic plasticity.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.