We propose a novel mask-guided method for missing value imputation, effetively validated on 25 datasets. By considering feature-wise and sample-wise correlations with novel modules, M³-Impute achieves best performance compared with SOTA methods.

We design a distributed GNN training framework that does not require storing the entire graph in memory for graph partitioning while enabling training GNN models on large graphs under limited resources.

We propose a context-free path recovery model and a novel online turn detection algorithm to generate path description from IMU signals.

We propose an MLP-mixer with novel patching technique and residual loss to process multivariate time series data and make forecasting accurately.

We design a floor identification system for crowdsourced RF signals in a building using only one labeled data sample from the bottom floor.

We propose a simple yet fast and effective partial convolution (PConv), as well as a latency-efficient family of architectures called FasterNet.

We propose an efficient bipartite graph sample and aggregate mechanism (BiSAGE) to process ambient RF signals for geofencing applications such as elderly care and pandemic control, etc.

TVConv works more computation-efficient than regular convolution when dealing with layout-specific tasks, e.g., face recognition.

We model the RF signals crowdsourced within a building as a bipartite graph and process the graph with an extended version of LINE (E-LINE) for floor identification tasks.

We propose PRID and PRID-lite to enable proximity detection using Bluetooth low energy (BLE) signals on commercial hardware and edge devices.

SignatureHome is the prototype system for home quarantine of COVID-19 in Hong Kong. It mainly relies on the associated IP address and ambient WiFi signals.

We analyse location errors for different localization systems theoretically and propose an optimized fusion scheme for different systems.