对生物医学图像进行操纵以虚假陈述实验结果,困扰着生物医学界。对该问题的最新兴趣导致了数据集和相关任务的策划,以促进生物医学法医方法的发展。其中,最大的操纵检测任务侧重于检测图像之间的重复区域。基于自然图像训练的法医模型的传统计算机视觉并非旨在克服生物医学图像带来的挑战。我们提出了一个多尺度重叠检测模型,以检测重复的图像区域。我们的模型的结构是从层次上找到重复,以减少补丁操作的数量。它总体上和多个生物医学图像类别都达到了最先进的性能。
translated by 谷歌翻译
Causal representation learning has been proposed to encode relationships between factors presented in the high dimensional data. However, existing methods suffer from merely using a large amount of labeled data and ignore the fact that samples generated by the same causal mechanism follow the same causal relationships. In this paper, we seek to explore such information by leveraging do-operation to reduce supervision strength. We propose a framework that implements do-operation by swapping latent cause and effect factors encoded from a pair of inputs. Moreover, we also identify the inadequacy of existing causal representation metrics empirically and theoretically and introduce new metrics for better evaluation. Experiments conducted on both synthetic and real datasets demonstrate the superiorities of our method compared with state-of-the-art methods.
translated by 谷歌翻译
在本文中,我们介绍了一种新颖的非线性激活功能,可以在神经网络的嵌入空间中自发地引起类紧凑性和正则化。该功能是被称为圆顶的镜像指数术语的差异。该函数的基本形式可以将SIGMOID或双曲线切线功能替换为二进制分类问题的输出激活功能。该功能也可以扩展到多级分类的情况,并用作标准SoftMax函数的替代方案。还可以进一步推广以采取适合网络的中间层的更灵活的形状。我们经验证明了该功能的属性。我们还显示使用该功能的模型对抗对抗攻击表现出额外的鲁棒性。
translated by 谷歌翻译
With the booming deployment of Internet of Things, health monitoring applications have gradually prospered. Within the recent COVID-19 pandemic situation, interest in permanent remote health monitoring solutions has raised, targeting to reduce contact and preserve the limited medical resources. Among the technological methods to realize efficient remote health monitoring, federated learning (FL) has drawn particular attention due to its robustness in preserving data privacy. However, FL can yield to high communication costs, due to frequent transmissions between the FL server and clients. To tackle this problem, we propose in this paper a communication-efficient federated learning (CEFL) framework that involves clients clustering and transfer learning. First, we propose to group clients through the calculation of similarity factors, based on the neural networks characteristics. Then, a representative client in each cluster is selected to be the leader of the cluster. Differently from the conventional FL, our method performs FL training only among the cluster leaders. Subsequently, transfer learning is adopted by the leader to update its cluster members with the trained FL model. Finally, each member fine-tunes the received model with its own data. To further reduce the communication costs, we opt for a partial-layer FL aggregation approach. This method suggests partially updating the neural network model rather than fully. Through experiments, we show that CEFL can save up to to 98.45% in communication costs while conceding less than 3% in accuracy loss, when compared to the conventional FL. Finally, CEFL demonstrates a high accuracy for clients with small or unbalanced datasets.
translated by 谷歌翻译
Batch Normalization (BN) is an important preprocessing step to many deep learning applications. Since it is a data-dependent process, for some homogeneous datasets it is a redundant or even a performance-degrading process. In this paper, we propose an early-stage feasibility assessment method for estimating the benefits of applying BN on the given data batches. The proposed method uses a novel threshold-based approach to classify the training data batches into two sets according to their need for normalization. The need for normalization is decided based on the feature heterogeneity of the considered batch. The proposed approach is a pre-training processing, which implies no training overhead. The evaluation results show that the proposed approach achieves better performance mostly in small batch sizes than the traditional BN using MNIST, Fashion-MNIST, CIFAR-10, and CIFAR-100 datasets. Additionally, the network stability is increased by reducing the occurrence of internal variable transformation.
translated by 谷歌翻译
表示解开是表示有利于各种下游任务的代表性学习的重要目标。为了实现这一目标,已经开发了许多无监督的学习表示方法。但是,事实证明,没有使用任何监督信号的培训过程就不足以进行分解表示。因此,我们提出了一种新型的弱监督训练方法,称为SW-VAE,该方法通过使用数据集的生成因子,将成对的输入观测值作为监督信号。此外,我们引入了策略,以逐渐增加训练过程中的学习难度,以使训练过程平滑。如多个数据集所示,我们的模型对表示解散任务的最新方法(SOTA)方法显示出显着改善。
translated by 谷歌翻译
我们提出语言学家,这是一种通过微调Alexatm 5B生成带注释数据的方法,用于生成意图分类和插槽标记(IC+ST),这是一种5亿参数的多语言序列到序列(SEQ2SEQ)模型,在灵活的指令上迅速的。在SNIP数据集的10次新颖意图设置中,语言学家超过了最新的方法(反向翻译和示例外推),可以通过宽阔的边距,显示出IC回忆中+1.9点的目标意图的绝对改善ST F1分数和+2.5分。在MATIS ++数据集的零击跨语言设置中,语言学家表现出强大的机器翻译基线,插槽对齐的基线是+4.14的+4.14点在6个语言上绝对在ST F1分数上,同时在IC上匹配IC的性能。最后,我们在用于对话代理IC+ST的内部大规模多语言数据集上验证了我们的结果,并显示了使用背面翻译,释义和插槽目录重新采样采样的基线的显着改进。据我们所知,我们是第一个展示大规模SEQ2SEQ模型的指导微调的人,以控制多语言意图和插槽标记的数据生成的输出。
translated by 谷歌翻译
散布和不变的表示是代表学习的两个关键目标,并且已经提出了许多方法来实现其中的一个。但是,这两个目标实际上是相互补充的,因此我们提出了一个框架,以同时完成两个目标。我们引入了一个弱监督的信号,以学习解开表示的表示,该表示由三个拆分组成,分别包含预测性,已知滋扰和未知的滋扰信息。此外,我们结合了对比度的实施表示不变性的方法。实验表明,所提出的方法在四个标准基准上优于最先进的方法(SOTA)方法,并表明该方法可以具有更好的对抗性防御能力,而没有对抗训练的其他方法。
translated by 谷歌翻译
由于视觉识别的社会影响一直受到审查,因此出现了几个受保护的平衡数据集,以解决不平衡数据集中的数据集偏差。但是,在面部属性分类中,数据集偏置既源于受保护的属性级别和面部属性级别,这使得构建多属性级别平衡的真实数据集使其具有挑战性。为了弥合差距,我们提出了一条有效的管道,以产生具有所需面部属性的高质量和足够的面部图像,并将原始数据集补充为两个级别的平衡数据集,从理论上讲,这在理论上满足了几个公平标准。我们方法的有效性在性别分类和面部属性分类方面得到了验证,通过将可比的任务性能作为原始数据集,并通过广泛的度量标准进行全面的公平评估,并进一步提高公平性。此外,我们的方法优于重采样和平衡的数据集构建来解决数据集偏差,以及解决任务偏置的模型模型。
translated by 谷歌翻译
由于电容层析成像(ECT)应用在几个工业领域的快速增长,因此从原始电容测量中开发出高质量但快速的图像重建方法的需求。深度学习是一种有效的非线性映射工具,用于复杂功能,在包括电断层扫描在内的许多领域都流行了。在本文中,我们提出了一个条件生成对抗网络(CGAN)模型,用于重建电容测量的ECT图像。 CGAN模型的初始图像是根据电容测量构建的。据我们所知,这是第一次以图像形式表示电容测量。我们创建了一个新的大规模ECT数据集,该数据集的320K合成图像测量对进行训练和测试所提出的模型。使用测试数据集,受污染的数据和流动模式评估所提出的CGAN-ECT模型的可行性和概括能力,这些数据集在训练阶段未暴露于模型。评估结果证明,与传统和其他基于学习的图像重建算法相比,提出的CGAN-ECT模型可以有效地创建更准确的ECT图像。 CGAN-ECT达到的平均图像相关系数超过99.3%,平均相对图像误差约为0.07。
translated by 谷歌翻译