12 September 2019 in IEEE Transactions on Signal Processing
In this paper, we design and implement a new on-line portfolio selection strategy based on reversion mechanism and weighted on-line learning. Our strategy, called “Gaussian Weighting Reversion” (GWR), improves the reversion estimator to form optimal portfolios and effectively overcomes the shortcomings of existing on-line portfolio selection strategies. Firstly, GWR uses Gaussian function to weight data in a sliding window to exploit the “time validity” of historical market data. It means that the more recent data are more valuable for market prediction than the earlier. Secondly, the self-learning for various sliding windows is created to make our strategy adaptive to different markets. In addition, double estimations are first proposed to be made at each time point, and the average of double estimations is obtained to alleviate the influence of noise and outliers. Extensive evaluation on six public datasets shows the advantages of our strategy compared with other nine competing strategies, including the state-of-the-art ones. Finally, the complexity analysis of GWR shows its availability in large-scale real-life online trading.
1 August 2017 in Review of Scientific Instruments
A flexible dual-mode proximity sensor has been designed and implemented, which is capable of combining capacitive-resistive detection in this paper. The capacitive type proximity sensor detecting is defined as mode-C, and the resistive type proximity sensor detecting is defined as mode-R. The characteristics of the proximity sensor are as follows: (1) the theoretical mode is developed which indicates that this proximity sensor can reflect proximity information accurately; (2) both sensing modes are vertically integrated into a sandwich-like chip with an 8 mm × 12 mm unit area. The thickness of a mode-R sensing material (graphene nanoplatelets) and mode-C dielectric (the mixture of carbon black and silicone rubber) is 1 mm and 2.5 mm, respectively; (3) for mode-R, the linearity of temperature-resistance curve can achieve 0.998 in the temperature range from C to C. And for mode-C, various materials can be successfully detected with fast response and high reversibility. Meanwhile, the study compensated the influence of object temperature to ensure mode-C properly works. A cooperative sensing test shows that R-C dual modes sense effectively which can enlarge the sensing distance compared with the single mode proximity sensor. The fabrication of this sensor is convenient, and the integrity of a flexible sandwich-like structure based on dual modes is beneficial to form arrays, which is suitable to be used in skin-like sensing applications.
16 September 2023 in Materials Horizons
A multi-step and multi-stage high-throughput materials search via ensemble machine learning is developed to carefully and comprehensively screen nearly 12 million all-inorganic perovskites with potential high photovoltaic performance.
31 October 2024 in IEEE Transactions on Automation Science and Engineering
Sensor-based human activity recognition is important in daily scenarios such as smart healthcare and homes due to its non-intrusive privacy and low cost advantages, but the problem of out-of-domain generalization caused by differences in focusing individuals and operating environments can lead to significant accuracy degradation on cross-person behavior recognition due to the inconsistent distributions of training and test data. To address the above problems, this paper proposes a new method, Multi-channel Time Series Decomposition Network (MTSDNet). Firstly, MTSDNet decomposes the original signal into a combination of multiple polynomials and trigonometric functions by the trainable parameterized time series decomposition to learn the low-rank representation of the original signal for improving the extraterritorial generalization ability of the model. Then, the different components obtained by the decomposition are classified layer by layer and the layer attention is used to aggregate components to obtain the final classification result. Extensive evaluation on DSADS, OPPORTUNITY, PAMAP2, UCIHAR and UniMib public datasets shows the advantages in classification accuracy and stability of our method compared with other competing strategies, including the state-of-the-art ones. And the visualization is conducted to reveal MTSDNet’s interpretability and layer-by-layer characteristics. Note to Practitioners —This paper is motivated by solving the problem of decreased classification accuracy caused by differences in human activity, such as the differences in walking patterns between young and elderly people. Models trained on young people’s data are difficult to accurately recognize the activities of the elderly. This is a problem of domain generalization, where the distribution differences between training and testing data require the model’s ability to generalize across domains. We propose a model based on time series decomposition, which helps the model learn universal features and effectively improve its generalization ability. This method can be further extended to time series tasks with varying domain distributions.
1 April 2015 in Journal of Polymer Engineering
A new method is reported for preparing poly(butylene terephthalate) (PBT)/glass fiber (GF)-silicon carbide (SiC) composites. GFs were coated with SiC particles firstly and then the treated GFs (GF-SiC) were mixed with PBT. Scanning electron microscopy (SEM) images showed that some SiC particles dispersed on the surfaces of GFs, and other particles fell off from surfaces of GFs and dispersed around GFs after processing. The thermal conductivities of composites are increased with increase of GF-SiC. At 30 wt% GF-SiC, the thermal conductivity of composites reached 0.6392 W/mK, which improved nearly 160% compared to that of neat polymer. Differential scanning calorimetry (DSC) results indicated that GF-SiC increased crystallinity of PBT compared with that of neat PBT and had no obvious influence on the melting temperature of PBT. The crystallization temperature (Tc) and glass transition temperature of all samples shifted to a higher temperature. The addition of GF-SiC can improve the thermal stability of PBT/GF-SiC composites obviously. Dynamic mechanical properties and dielectric properties were also discussed in this paper.
18 March 2025 in IEEE Sensors Journal
In sensor-based human activity recognition, data from different people are usually assumed to be identically independently distributed. However, an activity may vary from person to person so that a model trained on one person may perform terribly on other people because of different distributions, AKA domain shift. To solve the problem, we propose Domain Specific Mixture of Experts (DSME) especially for generalizing sensor-based human activity recognition. Firstly, DSME enables several experts to learn from their responsible source domains, respectively. Secondly, a domain-sniff module is used to extract features from each source domain to learn weights for experts, which manages to make classification more accurate. Finally, extensive experiments on four public datasets show that DSME gains 0.82%, 4.07%, 1.56% and 0.25% F1 score improvement compared with the-state-of-art baselines in DSA, PAMAP2, OPP and UCIHAR datasets. Besides, visualization analysis is also conducted to further prove that DSME can identify the discrepancy among training domains and then relieve domain shift in its best way.