A5049552955- Distributed and Parallel Computing Systems
- Advanced Data Storage Technologies
- Particle Detector Development and Performance
- Scientific Computing and Data Management
- Particle physics theoretical and experimental studies
- Advanced Image and Video Retrieval Techniques
- Library Science and Information Systems
- Image Processing and 3D Reconstruction
- Mathematics, Computing, and Information Processing
- Remote-Sensing Image Classification
- Healthcare Policy and Management
- Domain Adaptation and Few-Shot Learning
- Air Quality and Health Impacts
- Porphyrin and Phthalocyanine Chemistry
- Geographic Information Systems Studies
- Data Visualization and Analytics
- Speech and dialogue systems
- Atmospheric aerosols and clouds
- Advanced Vision and Imaging
- Neural Networks and Applications
- Data Mining Algorithms and Applications
- Ga2O3 and related materials
- Face and Expression Recognition
- Context-Aware Activity Recognition Systems
- Stochastic Gradient Optimization Techniques
Beijing Normal University - Hong Kong Baptist University United International College
2025
Microsoft Research Asia (China)
2025
Flatiron Health (United States)
2024
Xi'an Jiaotong University
2022-2024
Argonne National Laboratory
2010-2016
In recent years, dialogue generation models have emerged as a prominent area of research, crucial role in the domain natural language processing, garner significant attention from academic community. Existing predominantly emphasize human-computer interaction, however, various context-sensitive issues should not be overlooked inherent to process. Consequently, this paper aims summarize and categorize these processing issues. Firstly, based on extant literature, an overview current landscape...
Modern scientific experiments collect vast amounts of data that must be catalogued to meet multiple use cases and search criteria. In particular, high-energy physics currently in operation produce several billion events per year. A database with the references files including each event every stage processing is necessary order retrieve selected from storage systems. The ATLAS EventIndex project studying best way store information using modern technologies (Hadoop, HBase etc.) allow saving...
Abstract. Chronic high surface ozone (O3) levels and the increasing sulfur oxides (SOx = SO2+SO4) ambient concentrations over South Coast (SC) other areas of California (CA) are affected by both local emissions long-range transport. In this paper, multi-scale tracer, full-chemistry adjoint simulations using STEM atmospheric chemistry model conducted to assess contribution emission sourcesto SC O3 evaluate impacts transported on budgetduring ARCTAS-CARB experiment period in 2008. Sensitivity...
In the ATLAS experiment, a system called COMA (Conditions/Configuration Metadata for ATLAS), has been developed to make globally important run-level metadata more readily accessible. It is based on relational database storing directly extracted, refined, reduced, and derived information from system-specific sources as well non-database sources. This facilitates variety of unique dynamic interfaces provides enhance functionality other systems.
ATLAS has developed and deployed event-level selection services based upon event metadata records ("TAGS") supporting file database technology. These allow physicists to extract events that satisfy their predicates from any stage of data processing use them as input later analyses. One component these is a web-based Event-Level Selection Service Interface (ELSSI). ELSSI supports by integrating run-level metadata, luminosity-block-level (e.g., detector status quality information),...
By "intelligently" fuse the complementary information across different views, multi-view learning is able to improve performance of classification task. In this work, we extend bottleneck principle supervised scenario and use recently proposed matrix-based Rényi's α-order entropy functional optimize resulting objective directly, without necessity variational approximation or adversarial training. Empirical results in both synthetic real-world datasets suggest that our method enjoys improved...
In the ATLAS experiment, Tag Data, or short TAG, are event-level metadata -thumbnail information about events to support efficient identification and selection of interest a given analysis. TAG quantities range from detector status trigger basic physics quantities, e. g. number loose electrons candidates kinematic for limited these sorted by their transverse momentum. The average size per event is around 1kB, which factor 100 smaller than Analysis Object Data (AOD) used TAGs primarily...
ATLAS maintains a rich corpus of event-by-event information that provides global view the billions events collaboration has measured or simulated, along with sufficient auxiliary to navigate and retrieve data for any event at production processing stage. This unique resource been employed range purposes, from monitoring, statistics, anomaly detection, integrity checking, picking, subset selection, sample extraction. Recent years data-taking provide foundation assessment how this not used in...
This paper tackles two significant challenges in incomplete multiview learning. The first issue is how to comprehensively extract both consistent and view-specific information from data. second concerns eliminating irrelevant that either exists the original data or arises when imputing missing views. To address these issues, we introduce an information-theoretic framework tailored for Inspired by concept of common theory, our precisely defines what constitutes across We employ...
The ATLAS event-level metadata infrastructure supports applications that range from data quality monitoring, anomaly detection, and fast physics monitoring to selection navigation file-resident event at any processing stage, raw through analysis object data, in globally distributed analysis. A central component of the is a TAG database, which contains records for all events, real simulated. This resource offers unique global view provides an opportunity, not only stream-style mining but also...
We introduce an innovative and mathematically rigorous definition for computing common information from multi-view data, drawing inspiration G\'acs-K\"orner in theory. Leveraging this definition, we develop a novel supervised learning framework to capture both unique information. By explicitly minimizing total correlation term, the extracted each view are forced be independent of other, which, turn, theoretically guarantees effectiveness our framework. To estimate information-theoretic...
Large Language Models (LLMs) demonstrate exceptional performance in textual understanding and tabular reasoning tasks. However, their ability to comprehend analyze hybrid text, containing data, remains unexplored. The text often appears the form of long documents (HLDs), which far exceed token limit LLMs. Consequently, we apply an Automated Information Extraction framework (AIE) enable LLMs process HLDs carry out experiments analyse four important aspects information extraction from HLDs....
The ATLAS detector at the LHC takes data 200–500 Hz for several months per year accumulating billions of events hundreds physics analyses. TAGs are event-level metadata allowing a quick search interesting based on selection criteria defined by user. They stored in file-based format as well relational databases. overall TAG system architecture encompasses range interconnected services that provide functionality required use cases such event selection, display, extraction and skimming....
Femtosecond x-ray studies of 2D transition metal dichalcogenide films reveal ultrafast in-plane and out-of-plane responses, including compression the lattice spacing, structure factor modulations, dynamics occurring on few picosecond time-scales.