In this work, we propose several attention formulations for multivariate sequence data. We build on top of the recently introduced 2D-Attention and reformulate the attention learning methodology by quantifying the relevance of feature/temporal dimensions through latent spaces based on self-attention rather than learning them directly. In addition, we propose a joint feature-temporal attention mechanism that learns a joint 2D attention mask highlighting relevant information without treating feature and temporal representations independently. The proposed approaches can be used in various architectures and we specifically evaluate their application together with Neural Bag of Features feature extraction module. Experiments on several sequence data analysis tasks show the improved performance yielded by our approach compared to standard methods.
Bag of Instances Aggregation Boosts Self-supervised Distillation
Shanghai Jiao Tong University(上海交通大学);Huawei Inc.(华为公司);Huazhong University of Science & Technology(华中科技大学)
自监督学习的最新进展经历了显着的进步,特别是对于基于对比学习的方法,它将每个图像及其增强视为一个单独的类,并试图将它们与所有其他图像区分开来。然而,由于大量的样本,这种借口任务本质上遭受缓慢的收敛和难以优化。这对于小规模模型尤其如此,我们发现与监督模型相比,性能急剧下降。在本文中,我们提出了一个简单而有效的无监督学习的蒸馏策略。其亮点在于,相似样本之间的关系很重要,可以无缝地转移到学生身上,以提高成绩。我们的方法,被称为BINGO,是Bag of InstaNces aggregatiOn的缩写,旨在将教师学到的关系转移到学生身上。这里的实例包表示教师构建的一组相似样本,并分组在一个包中,蒸馏的目标是聚合学生关于包中实例的紧凑表示。值得注意的是,BINGO在小规模模型上实现了新的最先进性能,即,在ImageNet上分别使用ResNet-18和ResNet-34作为主干进行线性评估,获得了65.5%和68.9%的top-1准确率,显著超过了基线(52.5%和57.4%的top-1准确率)。该代码可在https://github.com/haohang96/bingo上获得。
Probabilistic smooth attention for deep multiple instance learning in medical imaging
Department of Computer Science and AI, University of Granada(计算机科学与人工智能系,格拉纳达大学);Department of Statistics and Operations Research, University of Granada(统计与运筹学系,格拉纳达大学);Research Centre for Information and Communication Technologies (CITIC), University of Granada(信息与通信技术研究中心(CITIC),格拉纳达大学);Department of Electrical and Computer Engineering, Northwestern University(电气与计算机工程系,西北大学);Center for Computational Imaging and Signal Analytics in Medicine, Northwestern University(医学计算成像与信号分析中心,西北大学)
SGPMIL: Sparse Gaussian Process Multiple Instance Learning
National and Kapodistrian University of Athens(希腊国家与卡波迪斯蒂亚诺斯大学);ÉTS Montréal(蒙特利尔ÉTS);Archimedes, Athena Research Center(阿提卡研究中心);CentraleSupélec, Université Paris-Saclay(巴黎萨克雷大学中央理工-supélec)
LadderMIL: Multiple Instance Learning with Coarse-to-Fine Self-Distillation
Institute for Regeneration and Repair, University of Edinburgh(再生与修复研究所,爱丁堡大学);School of Informatics, University of Edinburgh(信息学院,爱丁堡大学);Indica Labs(Indica实验室);School of Medicine, University of St Andrews(医学学院,圣安德鲁斯大学);Department of Basic Pathology, National Defense Medical College(基础病理学部门,国家防卫医疗学院)
Bag of Image Patch Embedding Behind the Success of Self-Supervised Learning
Meta AI - FAIR(Meta人工智能基础研究实验室);Center for Data Science, New York University(纽约大学数据科学中心);Inria(法国国家信息与自动化研究所);École normale supérieure(巴黎高等师范学院);CNRS(法国国家科学研究中心);PSL Research University(巴黎文理研究大学);Redwood Center(红杉中心);UC Berkeley(加州大学伯克利分校);Courant Institute(柯朗研究所)
Bag of Recurrence Patterns Representation for Time-Series Classification
The SPIN center, École Nationale Supérieure des Mines de Saint-Étienne(圣艾蒂安国立高等矿业学校SPIN中心);CREATIS center, INSA-Lyon(里昂国立应用科学学院CREATIS中心);University of Lyon(里昂大学)
时间序列分类在模式识别中引起了广泛的关注,因为金融和健康信息学等不同领域的应用都涉及到时间序列信号。特征袋模型(BoF)通过根据数据学习词典中“特征词”的频率来总结信号,在TSC任务中取得了很大的成功。本文提出了嵌入递归图(RP),一种可视化技术的动态系统分析,在TSC的BoF模型。传统的BoF方法从一维信号片段中提取特征,本文使用RP将时间序列转换为二维纹理图像,然后对其应用BoF。时间序列的图像表示使我们能够探索不同的视觉描述符,不适用于1D信号,并将TSC任务作为一个纹理识别问题。在UCI时间序列分类存档上的实验结果表明,与现有的BoF模型和最先进的算法相比,所提出的Bag of Recurrence patterns(BoR)的准确性得到了显着提高。
WeakSupCon: Weakly Supervised Contrastive Learning for Encoder Pre-training
Department of Electrical and Computer Engineering, University of Utah, Salt Lake City, UT, USA(电气与计算机工程系,犹他大学);Scientific Computing and Imaging Institute, University of Utah, SLC, UT, USA(科学计算与成像研究所,犹他大学);Department of Pathology, University of Utah, SLC, UT, USA(病理学系,犹他大学)
A novel multiple instance learning framework for COVID-19 severity assessment via data augmentation and self-supervised learning
Nanjing University(南京大学);National Institute of Healthcare Data Science, Nanjing University(南京大学医疗健康数据科学国家研究院);Southeast University(东南大学);Shanghai United Imaging Intelligence Co., Ltd.(上海联影智能科技有限公司);ShanghaiTech University(上海科技大学);Korea University(高丽大学);Central South University(中南大学);The Second Xiangya Hospital, Central South University(中南大学湘雅二医院);Hangzhou First People’s Hospital, Zhejiang University School of Medicine(浙江大学医学院附属杭州市第一人民医院)