Recent years have witnessed the success of deep learning methods in human activity recognition (HAR). The longstanding shortage labeled data inherently calls for a plethora semisupervised methods, and one most challenging common issues with is imbalanced distribution over classes. Although problem has long existed broad real-world HAR applications, it rarely explored literature. In this paper, we propose model from multimodal wearable sensory data. We aim to address not only challenges...

Brain-computer interface (BCI) is a system empowering humans to communicate with or control the outside world exclusively brain intentions. Electroencephalography (EEG)-based BCI one of promising solutions due its convenient and portable instruments. Despite extensive research EEG in recent years, it still challenging interpret signals effectively nature noise difficulties capturing inconspicuous relations between specific activities. Most existing works either only consider as chain-like...

The vast proliferation of sensor devices and Internet Things enables the applications sensor-based activity recognition. However, there exist substantial challenges that could influence performance recognition system in practical scenarios. Recently, as deep learning has demonstrated its effectiveness many areas, plenty methods have been investigated to address In this study, we present a survey state-of-the-art for human We first introduce multi-modality sensory data provide information...

The electroencephalogram (EEG) signal is a medium to realize brain-computer interface (BCI) system due its zero clinical risk and portable acquisition devices. Current EEG-based BCI research usually requires subject-specific adaptation step before can be employed by new user. In contrast, the subject-independent scenario, where well trained model directly applied users without precalibration, particularly desired. Considering this critical gap, focus in letter developing an effective EEG...

Motor imagery classification from EEG signals is essential for motor rehabilitation with a Brain-Computer Interface (BCI). Most current works on this issue require subject-specific adaptation step before applied to new user. Thus the research of directly extending pre-trained model users particularly desired and indispensable. As brain dynamics fluctuate considerably across different subjects, it challenging design practical hand-crafted features based prior knowledge. Regarding gap, paper...

An EEG-based Brain-Computer Interface (BCI) is a system that enables user to communicate with and intuitively control external devices solely using the user's intentions. Current BCI research usually involves subject-specific adaptation step before ready be employed by new user. However, subject-independent scenario, in which well-trained model can directly applied users without pre-calibration, particularly desirable yet rarely explored. Considering this critical gap, our focus paper...

Person identification technology recognizes individuals by exploiting their unique, measurable physiological and behavioral characteristics. However, the state-of-the-art person systems have been shown to be vulnerable, e.g., contact lenses can trick iris recognition fingerprint films deceive sensors. EEG (Electroencephalography)-based identification, which utilizes users' brainwave signals for offers a more resilient solution, draw lot of attention recently. accuracy still requires...

Trajectory similarity computation serves as a fundamental functionality of various spatial information applications. Although existing deep learning methods offer better efficiency and accuracy than non-learning solutions, they are still immature in trajectory embedding suffer from poor generality heavy preprocessing for training. Targeting these limitations, we propose novel framework named KGTS based on knowledge graph grid embedding, prompt unsupervised contrastive improved computation....

Offline Multi-Agent Reinforcement Learning (MARL) aims to learn optimal joint policies from pre-collected datasets without further interaction with the environment. Despite encouraging results achieved so far, we identify policy mismatch problem that arises employing diverse offline MARL datasets, a highly important ingredient for cooperative generalization yet largely overlooked by existing literature. Specifically, in case exhibit various policies, often occurs when individual actions...

Action advising endeavors to leverage supplementary guidance from expert teachers alleviate the issue of sampling inefficiency in Deep Reinforcement Learning (DRL). Previous agent-specific action methods are hindered by imperfections agent itself, while agent-agnostic approaches exhibit limited adaptability learning agent. In this study, we propose a novel framework called Agent-Aware trAining yet Agent-Agnostic Advising (A7) strike balance between two. The underlying concept A7 revolves...

Semi-supervised learning is crucial for alleviating labelling burdens in people-centric sensing. However, humangenerated data inherently suffer from distribution shift semi-supervised due to the diverse biological conditions and behavior patterns of humans. To address this problem, we propose a generic distributionally robust model on shifted data. Considering both discrepancy consistency between labeled unlabeled data, learn latent features that reduce person-specific preserve task-specific...

Multimodal features play a key role in wearable sensor based human activity recognition (HAR). Selecting the most salient adaptively is promising way to maximize effectiveness of multimodal data. In this regard, we propose "collect fully and select wisely" principle as well an interpretable parallel recurrent model with convolutional attentions improve performance. We first collect modality relations between each pair generate frames, then introduce attention mechanism prominent regions from...

Multi-modality is an important feature of sensor based activity recognition. In this work, we consider two inherent characteristics human activities, the spatially-temporally varying salience features and relations between activities corresponding body part motions. Based on these, propose a multi-agent spatial-temporal attention model. The mechanism helps intelligently select informative modalities their active periods. And multiple agents in proposed model represent with collective motions...

In image set classification, a considerable advance has been made by modeling the original sets second order statistics or linear subspace, which typically lie on Riemannian manifold. Specifically, they are Symmetric Positive Definite (SPD) manifold and Grassmann respectively, some algorithms have developed them for classification tasks. Motivated inability of existing methods to extract discriminatory features data manifolds, we propose novel algorithm combines multiple manifolds as sets....

Solar power is one of the most attractive green energy sources and plays a vital role in daily electricity supply. Since amount available solar uncontrollable, it essential to forecasting its availability so that plants can arrange supply advance. Global horizontal irradiance (GHI) key indicator power, highly accurate forecasts for which are required successfully integrate into grid. In this letter, we propose probabilistic transformer (ProSIT); novel deep learning model multihorizon GHI...

Biometric authentication involves various technologies to identify individuals by exploiting their unique, measurable physiological and behavioral characteristics. However, traditional biometric systems (e.g., face recognition, iris, retina, voice, fingerprint) are facing an increasing risk of being tricked tools such as anti-surveillance masks, contact lenses, vocoder, or fingerprint films. In this paper, we design a multimodal system named Deepkey, which uses both Electroencephalography...

Brain-Computer Interface (BCI) enables human to communicate with and intuitively control an external device through brain signals. Movement intention recognition paves the path for developing BCI applications. The current state-of-the-art in EEG based usually involves subject-specific adaptation before ready use. However, subject-independent scenario, which a well-trained model is directly applied new subjects without any pre-calibration, particularly desired yet rarely explored. In order...

Electroencephalogram (EEG)-based brain–computer interfaces (BCIs) allow users to use brain signals control external instruments, and movement intention detecting BCIs can aid in the rehabilitation of patients who have lost motor function. Existing studies this area mostly rely on cue-based data collection that facilitates sample labeling but introduces noise from cue stimuli; moreover, it requires extensive user training, cannot reflect real usage scenarios. In contrast, self-paced overcome...

Sharing ubiquitous mobile sensor data, especially physiological raises potential risks of leaking physical and demographic information that can be inferred from the time series data. Existing sensitive protection mechanisms depend on data transformation are effective only a particular attribute, together with usually requiring labels for training. Considering this gap, we propose novel user framework without using training dataset or being validated protecting one specific information. The...

Ubiquitous mobile sensors on human activity recognition pose the threat of leaking personal information that is implicitly contained within time-series sensor signals and can be extracted by attackers. Existing protective methods only support specific sensitive attributes require massive relevant ground truth for training, which unfavourable to users. To fill this gap, we propose a novel data transformation framework prohibiting leakage from data. The proposed transforms raw into new format,...

As brain dynamics fluctuate considerably across different subjects, it is challenging to design effective handcrafted features based on prior knowledge. Regarding this gap, paper proposes a Graph-based Convolutional Recurrent Attention Model (G-CRAM) explore EEG subjects for movement intention recognition. A graph structure first developed embed the positioning information of nodes, and then convolutional recurrent attention model learns from both spatial temporal dimensions adaptively...

With their rising adoption and integration into smart grids, heat pumps are becoming an increasingly important source of flexible energy. Heat pump flexibility can be utilized by using controllers to remotely manage operation while maintaining the temperature within predefined user comfort bounds. Traditional indoor modelling approaches require detailed information about deployment site, device specific parameters monitored data, making them inapplicable for majority deployments. This paper...