Neural graph collaborative filtering has received great recent attention due to its power of encoding the high-order neighborhood via the backbone graph neural networks. However, their robustness against noisy user-item interactions remains largely unexplored. Existing work on robust collaborative filtering mainly improves the robustness by denoising the graph structure, while recent progress in other fields has shown that directly adding adversarial perturbations in the embedding space can significantly improve the model robustness. In this work, we propose to improve the robustness of neural graph collaborative filtering via both denoising in the structure space and perturbing in the embedding space. Specifically, in the structure space, we measure the reliability of interactions and further use it to affect the message propagation process of the backbone graph neural networks; in the embedding space, we add in-distribution perturbations by mimicking the behavior of adversarial attacks and further combine it with contrastive learning to improve the performance. Extensive experiments have been conducted on four benchmark datasets to evaluate the effectiveness and efficiency of the proposed approach. The results demonstrate that the proposed approach outperforms the recent neural graph collaborative filtering methods especially when there are injected noisy interactions in the training data.

The altitude of a moving user is important context information for mobile technologies and applications. However, with the increasing pervasiveness of smartphones and abundant mobile applications, developers and users have gradually discovered that the height is more useful than altitude in many situations. The height is often a relative value, which is the vertical distance to the ground rather than the vertical distance to sea level, and we believe that it is useful in many applications, such as localization/navigation, sport/health and tourism/travel. In this paper, we first carried out a nation-wide online survey to confirm the desirability for the height information in mobile applications, and the result is positive. Then, we proposed HiMeter, an effective and accurate approach to calculating the height of the smartphone. HiMeter makes use of a low-power barometer on the smartphone and does not require GPS or back-server support. We concentrate on the vertical moving pattern of the user and designed several novel techniques, resulting in HiMeter not needing any reference points, and the complex process of calculating the absolute altitude can be avoided. The field studies show that HiMeter can achieve an accuracy of within 5 m in 90% of cases indoors and an accuracy of 10 m in 83% of cases outdoors. Compared to the existing works, HiMeter is more accurate and practical and is more suitable for usage in many mobile applications.

Hyperspectral images are high-dimensional data that capture detailed spectral information across a wide range of wavelengths, enabling the precise identification and analysis of different materials or objects. However, the high dimensionality of the data also introduces information redundancy and increases the computational overhead, making it necessary to perform band selection to retain the most discriminative and informative bands for the target task. Traditional band selection methods, such as ranking-based, searching-based, and clustering-based approaches, often rely on handcrafted features and heuristic rules, which fail to fully exploit the latent information and complex spatial–spectral relationships in hyperspectral images. To address this issue, this paper proposes a two-stage unsupervised band selection method based on deep reinforcement learning. First, we performed noise estimation preprocessing to filter out bands with high noise levels to reduce the interference in the agent’s learning process. Then, the band selection problem was formulated as a Markov Decision Process (MDP), where the agent learned an optimal band selection strategy through interactions with the environment. In the design of the reward function, the Optimal Index Factor (OIF) was introduced as the evaluation metric to encourage the agent to select bands with high information content and low redundancy, and thereby improve the efficiency and quality of the selection process. Experimental results on three hyperspectral datasets demonstrated that the proposed method could effectively improve the performance of the hyperspectral image band selection.

This paper presents the design and implementation of BTrack, a new approach for location tracking on mountain roads. It is as accurate as the GPS on common smartphones, but costs much less energy. To design such a location system, we carried out a nation-wide online survey to confirm the desirability for low power location tracking system, and the result is positive. Then, we proposed BTrack, and it makes two key technical contributions. The first is to propose a dynamic transition matrix Hidden Markov Model to combine the barometer and accelerometer reading hints for estimating the location of the user. The second is to design some novel techniques for parameter estimation, and proposed an adaptive algorithm to reduce the computational complexity. The field studies show that the accuracy of BTrack is no worse than GPS in 56% cases, meanwhile, the energy consumption is only about 30%. Compared to the existing works, BTrack is more suitable for location tracking on mountain roads.

The development of the Internet of Things (IoT) has accelerated research in indoor navigation systems, a majority of which rely on adequate wireless signals and sources. Nonetheless, deploying such a system requires periodic site-survey, which is time consuming and labor intensive. To address this issue, in this paper we present Canoe, an indoor navigation system that considers shopping mall scenarios. In our system, we do not assume any prior knowledge, such as floor-plan or the shop locations, access point placement or power settings, historical RSS measurements or fingerprints, etc. Instead, Canoe requires only that the shop owners collect and publish RSS values at the entrances of their shops and can direct a consumer to any of these shops by comparing the observed RSS values. The locations of the consumers and the shops are estimated using maximum likelihood estimation. In doing this, the direction of the target shop relative to the current orientation of the consumer can be precisely computed, such that the direction that a consumer should move can be determined. We have conducted extensive simulations using a real-world dataset. Our experiments in a real shopping mall demonstrate that if 50% of the shops publish their RSS values, Canoe can precisely navigate a consumer within 30 s, with an error rate below 9%.

This paper presents the design and implementation of a Location data collecting system only using the barometer sensor on the smartphone. It is energy efficient because of using low-power barometer sensor readings to infer the location, and it protects user privacy by providing the non-sensitive location data. To design such a location data collecting system, we make some key technical contributions, 1) a curve fitting based solution to remove the barometer sensor reading noise caused by weather change, 2) a deep learning algorithm to detect user moving activities based on the Restricted Boltzmann Machine, and 3) a clustering based extraction algorithm for signatures of different locations. The field studies show that SELoc provides user daily locations with an accuracy of 85%, meanwhile, the average energy consumption is only about 22% compared to GPS.