核心框架HeteroWear
,然后下一句We present HeteroWear...这部分是介绍框架的。然后技术架构设计
,然后HeteroWear separates...这部分。然后训练策略
,然后Training combines...这部分。然后实验评估
,然后Under leave-one-dataset-out...这部分,还有Held-out-device...还有Under 50% test-time...。然后消融分析
,然后Ablations show...这部分。不对,等下要准确,不要自己加内容?不对,标题的措辞不能改变原文内容哦?哦不对,规则是不能增删改内容,那h2/h3的标题是划分小节,内容都是原文的,不能改原文的文字,所以h2/h3的标题应该是用原文里的关键表述?不对,等下,不对,规则4是保留原文所有信息,不得增删改内容,所以不能自己编造标题内容?哦,不对,等下,我想想,比如,原文的结构和内容,首先,第一个自然段(背景)用p?不对,或者,正确的拆分: 首先,开头的背景部分是一个p?不对,再理清楚原文的内容分层: 1. 研究背景与现存问题:Self-supervised learning for wearable sensing is usually confined to a single dataset or device family, leaving open whether representations transfer across heterogeneous sensor configurations and whether gains come from pretraining rather than architecture or richer sensor input. 2. 提出的框架介绍:We present HeteroWear, a sensor-aware self-supervised framework pretrained on an eight-dataset wearable corpus (173 subjects, 358,261 windows) spanning inertial and physiological signals across body locations, devices, and sampling rates. 3. 技术架构:HeteroWear separates within-sensor temporal modeling from cross-sensor relational modeling using sensor-context embeddings, temporal and cross-sensor Transformers, and latent cross-attention fusion. 4. 训练方法:Training combines structured sensor dropout, cross-configuration consistency, and per-axis masked sensor modeling; missing sensors are omitted rather than reconstructed. 5. 实验结果(基准评估):Under leave-one-dataset-out pretraining and subject-disjoint evaluation on PAMAP2, WISDM, HHAR, HARTH, and UCI-HAR, HeteroWear improves mean macro-F1 from 0.688 to 0.724 over the identical architecture trained from scratch, with the largest gain on HHAR (0.823 versus 0.741), and achieves the highest macro-F1 on four of five benchmarks relative to single-accelerometer supervised baselines. 6. 跨设备评估:Held-out-device evaluation yields a mean +0.066 macro-F1 improvement. 7. 通道缺失鲁棒性评估:Under 50% test-time channel removal, HeteroWear retains 73.8% of full-channel performance versus 64.1% for the same architecture trained from scratch without sensor dropout. 8. 消融结果:Ablations show that structured sensor dropout primarily drives robustness, while cross-configuration consistency and sensor-context encoding contribute to transfer. 不对,但是h2/h3的标题需要是合理的,而且不能改变原文内容,哦不对,等下,规则里说长文用h2/h3划分小节,那标题的内容应该是对小节内容的概括,但是不能增删改原文的信息?不对,规则4是保留原文所有信息,不得增删改内容,就是说原文的文字不能改,不能加原文没有的信息,也不能删掉原文的信息。那h2/h3的标题是结构划分,应该用原文里的关键词?比如: 首先,开头的背景部分用研究背景
?不对,会不会不符合?不对,等下,再看规则,规则3里的重要句子可以用strong,那重要的术语比如HeteroWear可以加strong?对哦,规则3说重要句子可用。 等下,重新来,正确的HTML结构: 首先,第一个部分是研究背景,作为第一个小节?研究背景与问题
Self-supervised learning for wearable sensing is usually confined to a single dataset or device family, leaving open whether representations transfer across heterogeneous sensor configurations and whether gains come from pretraining rather than architecture or richer sensor input.
然后下一个小节是框架介绍:核心框架HeteroWear
We present HeteroWear, a sensor-aware self-supervised framework pretrained on an eight-dataset wearable corpus (173 subjects, 358,261 windows) spanning inertial and physiological signals across body locations, devices, and sampling rates.
技术架构设计
HeteroWear separates within-sensor temporal modeling from cross-sensor relational modeling using sensor-context embeddings, temporal and cross-sensor Transformers, and latent cross-attention fusion.
训练策略
Training combines structured sensor dropout, cross-configuration consistency, and per-axis masked sensor modeling; missing sensors are omitted rather than reconstructed.
然后是实验评估的小节:实验评估结果
Under leave-one-dataset-out pretraining and subject-disjoint evaluation on PAMAP2, WISDM, HHAR, HARTH, and UCI-HAR, HeteroWear improves mean macro-F1 from 0.688 to 0.724 over the identical architecture trained from scratch, with the largest gain on HHAR (0.823 versus 0.741), and achieves the highest macro-F1 on four of five benchmarks relative to single-accelerometer supervised baselines.
Held-out-device evaluation yields a mean +0.066 macro-F1 improvement.
Under 50% test-time channel removal, HeteroWear retains 73.8% of full-channel performance versus 64.1% for the same architecture trained from scratch without sensor dropout.
然后是消融分析的小节:消融分析结论
Ablations show that structured sensor dropout primarily drives robustness, while cross-configuration consistency and sensor-context encoding contribute to transfer.
等下,会不会标题是我自己加的内容?不对,规则说不得增删改内容,那标题的概括算不算增改?哦,不对,等下,原文是一整段的学术文本,可能不需要硬加标题?不对,规则说长文用h2/h3划分小节,那原文的内容是长文,所以需要划分。等下,再仔细看规则4:“保留原文所有信息,不得增删改内容”,意思是原文的文字内容不能改动,不能增加原文没有的事实信息,但是结构划分的标签是排版用的,标题是对各部分内容的提炼,只要是准确的,不改变原意,不增加信息,是可以的?不对,或者,有没有可能,不需要加标题?不对,规则2说“长文用