Deep Neural Networks are Easily Fooled: 这篇英文论文的翻译谁有

Deep Neural Networks are Easily Fooled: 这篇英文论文的翻译谁有

Deep NeuralNetworks are Easily Fooled: High Confidence Predictions for UnrecognizableImages
深层神经网络容易受骗上当:准确识别难以辨别的图像
Deep neuralnetworks (DNNs) have recently been achieving state-of-the-art performance on avariety of pattern-recognition tasks, most notably visual classificationproblems.
最近，深层神经网络系统（DNNs）实现了能在在多种模式下识别图像的先进的性能,尤其是在视觉分类问题上。
Given that DNNsare now able to classify objects in images with near-human-level performance,questions naturally arise as to what differences remain between computer andhuman vision.
如今，现款的DDNs能够以近似人类的视觉水平对图像进行分类，因此自然而然地，人们对于计算机和人类的视觉区别的分歧依然存在。
A recent study revealed that changing an image(e.g. of a lion) in a way imperceptible to humans can cause a DNN to label theimage as something else entirely (e.g. mislabeling a lion a library).
最近一项研究表明，用人类察觉不到的方式改变一个图像（比如一只狮子），人类看到图像以后察觉不出变化，而DDNs却能够识别出图像的变化并把它标记成一张完全不同的图像（就好比一只图书馆里贴错标签的狮子)。
Here we show arelated result: it is easy to produce images that are completely unrecognizableto humans, but that state-of-the-art DNNs believe to be recognizable objectswith 99.99% confidence (e.g. labeling with certainty that white noise static isa lion).
有一些相关的研究结果表明：人类显然不容易辨别图像，但目前最先进的DDNs却能以99.99%的准确度辨别物体(比如能从静态的白噪声中辨别出是狮子)。

Specifically, we take convolutional neural networks trained to perform well on either the ImageNet or MNIST datasets and then find images with evolutionary algorithms or gradient ascent that DNNs label with high confidence as belonging to each dataset class.
具体来说，我们的卷积神经网络在图像网络或MNIST的数据上的表现都很良好，然而接下来我们会发现，DDNs 能运用每个数据集类高度准确辨别那些经过进化算法和梯度上升的图像。

It is possible toproduce images totally unrecognizable to human eyes that DNNs believe with nearcertainty are familiar objects, which we call "fooling images" (moregenerally, fooling examples).
我们完全有可能制作出人类完全不能辨认出来的图像，但这些图像对于DDNs来说确实熟悉的可以辨认的物体，我们称之为“欺骗图像”(更普遍，更具迷惑性的图像)。
Our results shedlight on interesting differences between human vision and current DNNs, andraise questions about the generality of DNN computer vision.
我们的研究结果揭示了人类的视觉和目前DDNs之间有趣的差异，并提高了人们对计算机视觉的普遍性问题的关注。追问评论里将y输入函数后为什么上下重复追答and passing the y input into a sine function provides top-bottom repetition. Evolution determines
the topology, weights, and activation functions of each CPPN network in the population.
通过把字母y输入到一个正弦函数，系统因此能够循环运作。进化的数据结果决定了每个CPPN网络人口的拓扑，权重和激活功能。
(如果你上过计算机编程的课就知道计算机里面的编程是怎么运作的，类似循环的系统这样理解，专业的术语我也不太会）