A5018191388- Domain Adaptation and Few-Shot Learning
- Multimodal Machine Learning Applications
- Advanced Graph Neural Networks
- Air Quality Monitoring and Forecasting
- Fire effects on ecosystems
- Air Quality and Health Impacts
- COVID-19 diagnosis using AI
- Meteorological Phenomena and Simulations
- Recommender Systems and Techniques
- Flood Risk Assessment and Management
- Advanced Clustering Algorithms Research
- Advanced Image and Video Retrieval Techniques
- Machine Learning in Bioinformatics
- Protein Structure and Dynamics
- Millimeter-Wave Propagation and Modeling
- Remote-Sensing Image Classification
- Machine Learning and ELM
- Cancer-related molecular mechanisms research
- Atmospheric chemistry and aerosols
- Advanced Chemical Sensor Technologies
- Speech and Audio Processing
- Wireless Communication Networks Research
- Spam and Phishing Detection
- Impact of Light on Environment and Health
- Advanced MIMO Systems Optimization
Beijing Jiaotong University
2017-2025
Tsinghua University
2023-2025
Protein language models have shown remarkable success in learning biological information from protein sequences. However, most existing are limited by either autoencoding or autoregressive pre-training objectives, which makes them struggle to handle understanding and generation tasks concurrently. We propose a unified model, xTrimoPGLM, address these two types of simultaneously through an innovative framework. Our key technical contribution is exploration the compatibility potential for...
Computer vision systems have attracted much attention in recent years for use detecting surface defects on rails; however, accurate and efficient recognition of possible remains challenging due to the variations shown by also noise. This paper proposes a coarse-to-fine model (CTFM) identify at different scales. The works three scales from coarse fine: subimage level, region pixel level. At background subtraction exploits row consistency longitudinal direction, strongly filters defect-free...
Self-supervised learning (SSL) has recently achieved great success in mining the user-item interactions for collaborative filtering. As a major paradigm, contrastive (CL) based SSL helps address data sparsity Web platforms by contrasting embeddings between raw and augmented data. However, existing CL-based methods mostly focus on batch-wise way, failing to exploit potential regularity feature dimension. This leads redundant solutions during representation of users items. In this work, we...
Clustering of multipath components (MPCs) is an important aspect propagation channel modeling. When a time series measurements, based on movement transmitter and/or receiver, available, the temporal evolution MPCs can be used as basis for clustering. We present algorithm that bases clustering not only distance in delay/angle space, but also how similar their parameters are. Sample results obtained from vehicle-to-vehicle measurement campaign show good performance proposed algorithm.
Self-supervised learning (SSL) has recently achieved great success in mining the user-item interactions for collaborative filtering. As a major paradigm, contrastive (CL) based SSL helps address data sparsity Web platforms by contrasting embeddings between raw and augmented data. However, existing CL-based methods mostly focus on batch-wise way, failing to exploit potential regularity feature dimension. This leads redundant solutions during representation of users items. In this work, we...
Lightning as a natural phenomenon poses serious threats to human life, aviation and electrical infrastructures. prediction plays vital role in lightning disaster reduction. Existing methods, usually based on numerical weather models, rely parameterization schemes for forecasting. These however, have two drawbacks. Firstly, simulations of the models deviations space time domains, which introduces irreparable biases subsequent processes. Secondly, are designed manually by experts meteorology,...
Abstract Weather forecasting requires comprehensive analysis of a variety meteorological data. Recent decades have witnessed the advance weather observation and simulation technologies, triggering an explosion data which are collected from multiple sources (e.g., radar, automatic stations numerical prediction) usually characterized by spatiotemporal (ST) structure. As result, adequate exploition these multi‐source ST emerges as promising but challenging topic for forecasting. To address this...
Protein language models have shown remarkable success in learning biological information from protein sequences. However, most existing are limited by either autoencoding or autoregressive pre-training objectives, which makes them struggle to handle understanding and generation tasks concurrently. We propose a unified model, xTrimoPGLM, address these two types of simultaneously through an innovative framework. Our key technical contribution is exploration the compatibility potential for...
Accurate Remaining Useful Life (RUL) prediction is vital for effective prognostics in and the health management of industrial equipment, particularly under varying operational conditions. Existing approaches to multi-condition RUL often treat each working condition independently, failing effectively exploit cross-condition knowledge. To address this limitation, paper introduces MoEFormer, a novel framework that combines Mixture Encoders (MoE) with Transformer-based architecture achieve...
Multi-task model merging offers an efficient solution for integrating knowledge from multiple fine-tuned models, mitigating the significant computational and storage demands associated with multi-task training. As a key technique in this field, Task Arithmetic (TA) defines task vectors by subtracting pre-trained ($\theta_{\text{pre}}$) models parameter space, then adjusting weight between these $\theta_{\text{pre}}$ to balance task-generalized task-specific knowledge. Despite promising...
Graph Contrastive Learning (GCL), as a primary paradigm of graph self-supervised learning, spurs fruitful line research in tackling the data sparsity issue by maximizing consistency user/item embeddings between different augmented views with random perturbations. However, diversity, crucial metric for recommendation performance and user satisfaction, has received rather little attention. In fact, there exists challenging dilemma balancing accuracy diversity. To address these issues, we...
Accurate lightning forecast is significant for disaster prevention and reduction. However, the mainstream methods, which mainly rely on numerical simulations parameterizations, can hardly cope with spatiotemporal deviations. Meanwhile, rapid complex evolution of regions go beyond traditional extrapolation-based methods. In this work, we propose a data-driven neural network model hourly forecast, exploits both recent historical observations. The two kinds data complement each other play...
Radio channel modeling has been an important research topic, since the performance of any communication system depends on characteristics. So far, most existing clustering algorithms are conducted based multipath components (MPCs) extracted by using a high-resolution parameter estimation approach, e.g., SAGE or MUSIC, etc. However, approaches require prior information to extract MPCs. Moreover, usually result in relatively high complexity, and thus, clusters can only be identified offline...
Air pollution is growing ever more serious as a result of rising consumption energy and other natural resources. Generally, governmental static monitoring stations provide accurate air data, but they are sparsely distributed in the space. In contrast, microstations kind low-cost equipment can be densely though their accuracy relatively low. This article proposes deep calibration method (DeepCM) for sensors equipped microstations, which consists an encoder decoder. encoding stage, multilevel...
In mobile communications, wireless channel has been widely considered to be time-variant. To accurately model the time-variant channels, a tracking-based dynamic multipath components (MPCs) clustering algorithm is proposed in this letter. The tracking problem as total probability maximization estimation and solved by using Kuhn-Munkres algorithm. MPCs are further clustered based on results of tracking. validated simulated channels compared with distance-based method. Simulation show good performance
The dilemma between plasticity and stability arises as a common challenge for incremental learning. In contrast, the human memory system is able to remedy this owing its multilevel structure, which motivates us propose Bilevel Memory with Knowledge Projection (BMKP) BMKP decouples functions of learning remembering via bilevel-memory design: working responsible adaptively model learning, ensure plasticity; long-term in charge enduringly storing knowledge incorporated within learned model,...
Distributed clustering is emerging along with the advent of era big data. However, most existing established distributed methods focus on problems caused by a large amount data rather than dimension Consequently, they suffer "curse" dimensionality (e.g., poor performance and heavy network overhead) when high-dimensional (HD) are clustered. In this article, we propose algorithm, referred to as Local Density Subspace Clustering (LDSDC) cluster large-scale HD data, motivated idea that local...