Affinity-Based Hierarchical Learning of Dependent Concepts for Human Activity Recognition
LIPN-UMR CNRS 7030, Univ. Sorbonne Paris Nord(巴黎第十三大学(索邦巴黎北部大学)LIPN-UMR CNRS 7030实验室);Léonard de Vinci Pôle Universitaire, Research Center(达芬奇大学联盟研究中心)
In multi-class classification tasks, like human activity recognition, it is often assumed that classes are separable. In real applications, this assumption becomes strong and generates inconsistencies. Besides, the most commonly used approach is to learn classes one-by-one against the others. This computational simplification principle introduces strong inductive biases on the learned theories. In fact, the natural connections among some classes, and not others, deserve to be taken into account. In this paper, we show that the organization of overlapping classes (multiple inheritances) into hierarchies considerably improves classification performances. This is particularly true in the case of activity recognition tasks featured in the SHL dataset. After theoretically showing the exponential complexity of possible class hierarchies, we propose an approach based on transfer affinity among the classes to determine an optimal hierarchy for the learning process. Extensive experiments show improved performances and a reduction in the number of examples needed to learn.
HiF-DTA: Hierarchical Feature Learning Network for Drug-Target Affinity Prediction
School of Software Engineering Huazhong University of Science(软件工程学院 华中科技大学);Faculty of Data Science City University of Macau Macau, China(数据科学学院 澳门城市大学);Union Hospital, Tongji Medical College Huazhong University of Science(同济医学院附属医院 华中科技大学)
Structure-Aware Compound-Protein Affinity Prediction via Graph Neural Networks with Group Lasso Regularization
Department of Biostatistics & Health Data Science(生物统计学与健康数据科学系);Indiana University(印第安纳大学);Department of Computer Science(计算机科学系);Indiana University Bloomington(印第安纳大学布卢明顿分校);Division of Clinical Pharmacology(临床药理学部);Indiana University School of Medicine(印第安纳大学医学院);IUSM-Purdue TREAT-AD Center(IUSM-普渡大学TREAT-AD中心);Department of Medical and Molecular Genetics(医学与分子遗传学系)
CHARM: A Hierarchical Deep Learning Model for Classification of Complex Human Activities Using Motion Sensors
Meta Platforms, Inc.(元平台公司)
在本文中,我们报告了一个层次化的深度学习模型,用于使用运动传感器对复杂的人类活动进行分类。与用于基于事件的活动识别(例如计步、跌倒检测和手势识别)的传统人类活动识别(HAR)模型相比,我们将这种新的深度学习模型称为CHARM。复杂人类活动识别模型(Complex Human Activity Recognition Model),旨在识别由多个不同的低级别活动以非确定性顺序组成的高级别人类活动,例如准备膳食、家务和日常事务。CHARM不仅在平均准确度和F1分数方面在数量上优于高级活动识别的最先进的监督学习方法,而且还自动学习识别低级活动,例如操作手势和运动模式,而无需对此类活动进行任何显式标记。这为使用可穿戴传感器的人机交互(HMI)模态开辟了新的途径,其中用户可以选择将自动化任务与高级活动相关联,诸如控制家庭自动化(例如,机器人真空吸尘器、灯和恒温器)或在正确的时间呈现上下文相关的信息(例如,提醒、状态更新和天气/新闻报道)。此外,当仅使用高级活动标签进行训练时学习低级用户活动的能力可以为固有地难以标记的HAR任务的半监督学习铺平道路。
Stratified Transfer Learning for Cross-domain Activity Recognition
Beijing Key Laboratory of Mobile Computing and Pervasive Device Institute of Computing Technology, Chinese Academy of Sciences(中国科学院计算技术研究所移动计算与普适设备北京市重点实验室);University of Chinese Academy of Sciences(中国科学院大学);Institute of Information Technology, Hebei University of Economics and Business(河北经贸大学信息技术学院);Department of Computer Science, University of Illinois at Chicago(伊利诺伊大学芝加哥分校计算机科学系)
Advanced Global Wildfire Activity Modeling with Hierarchical Graph ODE
University of Science and Technology of China(中国科学技术大学);Tsinghua University(清华大学);Southwest Jiaotong University(西南交通大学);Beijing Jiaotong University(北京交通大学);Tencent(腾讯);Nanyang Technological University(南洋理工大学)
Probabilistic Structure Learning for EEG/MEG Source Imaging with Hierarchical Graph Prior
Massachusetts General Hospital(麻省总医院);Harvard Medical School(哈佛医学院);Picower Institute for Learning and Memory(皮考尔学习与记忆研究所);MIT(麻省理工学院);The University of Texas at Arlington(德克萨斯大学阿灵顿分校);The University of Texas at Dallas(德克萨斯大学达拉斯分校)
Leveraging Activity Recognition to Enable Protective Behavior Detection in Continuous Data
University College London(伦敦大学学院);Uppsala University(乌普萨拉大学);Nokia Bell Labs(诺基亚贝尔实验室);University of Cambridge(剑桥大学)
慢性疼痛(CP)患者在体力活动中表现出的保护性行为是了解其身体和情绪状态的关键。现有的自动保护行为检测(PBD)方法依赖于用户预定义的活动的预分割。然而,在现实生活中,人们随意地进行活动。因此,当这些活动对慢性疼痛患者造成困难时,应持续提供技术支持,并自动适应活动类型和保护行为的发生。因此,为了促进无处不在的CP管理,在连续数据上实现准确的PBD变得至关重要。在本文中,我们提出通过一种新的分层HAR-PBD架构将人类活动识别(HAR)与PBD集成在一起,该架构包括图卷积和长短期记忆(GC-LSTM)网络,并使用类平衡的焦点类别交叉熵(CFCC)损失来缓解类不平衡。通过使用CP患者数据集对该方法进行深入评估,我们发现HAR,GC-LSTM网络和CFCC损失的利用导致PBD性能相对于基线明显增加(宏观F1评分为0.81 vs. 0.66,精确召回曲线下面积(PR-AUC)为0.60 vs. 0.44)。最后,我们讨论了CP管理及其他方面的分层架构的可能用例。我们还讨论了目前的局限性和前进的道路。