arXiv subjects

Weiguang Chen

Publications and source records attributed to Weiguang Chen.

5 recordsLinked to original sources

arXiv2025

UniArray: Unified Spectral-Spatial Modeling for Array-Geometry-Agnostic Speech Separation

Array-geometry-agnostic speech separation (AGA-SS) aims to develop an effective separation method regardless of the microphone array geometry. Conventional methods rely on permutation-free operations, such as summation or attention mechanisms, to capture spatial information. However, these approaches often incur high computational costs or disrupt the effective use of spatial information during intra- and inter-channel interactions, leading to suboptimal performance. To address these issues, we propose UniArray, a novel approach that abandons the conventional interleaving manner. UniArray consists of three key components: a virtual microphone estimation (VME) module, a feature extraction and fusion module, and a hierarchical dual-path separator. The VME ensures robust performance across arrays with varying channel numbers. The feature extraction and fusion module leverages a spectral feature extraction module and a spatial dictionary learning (SDL) module to extract and fuse frequency-bin-level features, allowing the separator to focus on using the fused features. The hierarchical dual-path separator models feature dependencies along the time and frequency axes while maintaining computational efficiency. Experimental results show that UniArray outperforms state-of-the-art methods in SI-SDRi, WB-PESQ, NB-PESQ, and STOI across both seen and unseen array geometries.

cs.SD↗

arXiv2024

GAMMA: Graspability-Aware Mobile MAnipulation Policy Learning based on Online Grasping Pose Fusion

Mobile manipulation constitutes a fundamental task for robotic assistants and garners significant attention within the robotics community. A critical challenge inherent in mobile manipulation is the effective observation of the target while approaching it for grasping. In this work, we propose a graspability-aware mobile manipulation approach powered by an online grasping pose fusion framework that enables a temporally consistent grasping observation. Specifically, the predicted grasping poses are online organized to eliminate the redundant, outlier grasping poses, which can be encoded as a grasping pose observation state for reinforcement learning. Moreover, on-the-fly fusing the grasping poses enables a direct assessment of graspability, encompassing both the quantity and quality of grasping poses.

cs.RO↗

arXiv2020

Semi-Disentangled Representation Learning in Recommendation System

Disentangled representation has been widely explored in many fields due to its maximal compactness, interpretability and versatility. Recommendation system also needs disentanglement to make representation more explainable and general for downstream tasks. However, some challenges slow its broader application -- the lack of fine-grained labels and the complexity of user-item interactions. To alleviate these problems, we propose a Semi-Disentangled Representation Learning method (SDRL) based on autoencoders. SDRL divides each user/item embedding into two parts: the explainable and the unexplainable, so as to improve proper disentanglement while preserving complex information in representation. The explainable part consists of $internal\ block$ for individual-based features and $external\ block$ for interaction-based features. The unexplainable part is composed by $other\ block$ for other remaining information. Experimental results on three real-world datasets demonstrate that the proposed SDRL could not only effectively express user and item features but also improve the explainability and generality compared with existing representation methods.

cs.SI↗

arXiv2019

Band gap and band alignment prediction of nitride based semiconductors using machine learning

Nitride has been drawing much attention due to its wide range of applications in optoelectronics and remains plenty of room for materials design and discovery. Here, a large set of nitrides have been designed, with their band gap and alignment being studied by first-principles calculations combined with machine learning. Band gap and band offset against wurtzite GaN accurately calculated by the combination of screened hybrid functional of HSE and DFT-PBE were used to train and test machine learning models. After comparison among different techniques of machine learning, when elemental properties are taken as features, support vector regression (SVR) with radial kernel performs best for predicting both band gap and band offset with prediction root mean square error (RMSE) of 0.298 eV and 0.183 eV, respectively. The former is within HSE calculation uncertainty and the latter is small enough to provide reliable predictions. Additionally, when band gap calculated by DFT-PBE was added into the feature space, band gap prediction RMSE decreases to 0.099 eV. Through a feature engineering algorithm, elemental feature space based band gap prediction RMSE further drops by around 0.005 eV and the relative importance of elemental properties for band gap prediction was revealed. Finally, band gap and band offset of all designed nitrides were predicted and two trends were noticed that as the number of cation types increases, band gap tends to narrow down while band offset tends to go up. The predicted results will be a useful guidance for precise investigation on nitride engineering.

cond-mat.mtrl-sci↗

arXiv2016

Multiple Unpinned Dirac Points in Group-Va Single-layers with Phosphorene Structure

Emergent Dirac fermion states underlie many intriguing properties of graphene, and the search for them constitute one strong motivation to explore two-dimensional (2D) allotropes of other elements. Phosphorene, the ultrathin layers of black phosphorous, has been a subject of intense investigations recently, and it was found that other group-Va elements could also form 2D layers with similar puckered lattice structure. Here, by a close examination of their electronic band structure evolution, we discover two types of Dirac fermion states emerging in the low-energy spectrum. One pair of (type-I) Dirac points is sitting on high-symmetry lines, while two pairs of (type-II) Dirac points are located at generic $k$-points, with different anisotropic dispersions determined by the reduced symmetries at their locations. Such fully-unpinned (type-II) 2D Dirac points are discovered for the first time. In the absence of spin-orbit coupling, we find that each Dirac node is protected by the sublattice symmetry from gap opening, which is in turn ensured by any one of three point group symmetries. The spin-orbit coupling generally gaps the Dirac nodes, and for the type-I case, this drives the system into a quantum spin Hall insulator phase. We suggest possible ways to realize the unpinned Dirac points in strained phosphorene.

cond-mat.mtrl-sci↗