一家美国的跨国科技企业,致力于互联网搜索、云计算、广告技术等领域,由当时在斯坦福大学攻读理学博士的拉里·佩奇和谢尔盖·布林共同创建。创始之初,Google 官方的公司使命为「整合全球范围的信息,使人人皆可访问并从中受益」。 Google 开发并提供了大量基于互联网的产品与服务,其主要利润来自于 AdWords 等广告服务。

2004 年 8 月 19 日, 公司以「GOOG」为代码正式登陆纳斯达克交易所。

VIP内容

题目: Smooth Adversarial Training

摘要:

人们通常认为,网络不能兼具准确性和鲁棒性,获得鲁棒性意味着失去准确性。还普遍认为,除非扩大网络规模,否则网络架构元素对提高对抗性的健壮性影响不大。本文通过对对抗训练的仔细研究,提出了挑战这些共同信念的证据。主要观察结果是,广泛使用的ReLU激活功能由于其不平滑的特性而大大削弱了对抗训练。因此,提出了平滑对抗训练(SAT),在其中我们用ReLU平滑近似代替了ReLU,以加强对抗训练。SAT中平滑激活函数的目的是使它能够找到更难的对抗示例,并在对抗训练期间计算出更好的梯度更新。与标准的对抗训练相比,SAT提高了“free”的对抗鲁棒性,即准确性没有降低,计算成本也没有增加。例如,在不引入其他计算的情况下,SAT可将ResNet-50的鲁棒性从33.0%提高到42.3%,同时还将ImageNet的准确性提高0.9%。SAT在较大的网络上也能很好地工作:它可以帮助EfficientNet-L1在ImageNet上实现82.2%的准确性和58.6%的鲁棒性,在准确性和鲁棒性方面分别比以前的最新防御提高9.5%和11.6%。

成为VIP会员查看完整内容
0
12

最新论文

Google Trends is a tool that allows researchers to analyze the popularity of Google search queries across time and space. In a single request, users can obtain time series for up to 5 queries on a common scale, normalized to the range from 0 to 100 and rounded to integer precision. Despite the overall value of Google Trends, rounding causes major problems, to the extent that entirely uninformative, all-zero time series may be returned for unpopular queries when requested together with more popular queries. We address this issue by proposing Google Trends Anchor Bank (G-TAB), an efficient solution for the calibration of Google Trends data. Our method expresses the popularity of an arbitrary number of queries on a common scale without being compromised by rounding errors. The method proceeds in two phases. In the offline preprocessing phase, an "anchor bank" is constructed, a set of queries spanning the full spectrum of popularity, all calibrated against a common reference query by carefully chaining together multiple Google Trends requests. In the online deployment phase, any given search query is calibrated by performing an efficient binary search in the anchor bank. Each search step requires one Google Trends request, but few steps suffice, as we demonstrate in an empirical evaluation. We make our code publicly available as an easy-to-use library at https://github.com/epfl-dlab/GoogleTrendsAnchorBank.

0
0
下载
预览
Top