A5070531404- Recommender Systems and Techniques
- Advanced Data Compression Techniques
- Image Retrieval and Classification Techniques
- Image and Video Quality Assessment
- AI in cancer detection
- Image and Object Detection Techniques
- Network Traffic and Congestion Control
- Machine Learning and Data Classification
- Ferroelectric and Negative Capacitance Devices
- Wireless Communication Networks Research
- Caching and Content Delivery
- Parallel Computing and Optimization Techniques
- Mathematics, Computing, and Information Processing
- Neural dynamics and brain function
- Distributed and Parallel Computing Systems
- Advanced Computational Techniques and Applications
- Technology and Data Analysis
- IoT and Edge/Fog Computing
- Natural Language Processing Techniques
- Library Science and Information Systems
- Text and Document Classification Technologies
- IPv6, Mobility, Handover, Networks, Security
- Medical Image Segmentation Techniques
- Advanced Memory and Neural Computing
Korea Advanced Institute of Science and Technology
2025
Kuaishou (China)
2023
Beijing University of Posts and Telecommunications
2019
Click-through rate prediction is an essential task in industrial applications, such as online advertising. Recently deep learning based models have been proposed, which follow a similar Embedding\&MLP paradigm. In these methods large scale sparse input features are first mapped into low dimensional embedding vectors, and then transformed fixed-length vectors group-wise manner, finally concatenated together to fed multilayer perceptron (MLP) learn the nonlinear relations among features....
Increasing the size of embedding layers has shown to be effective in improving performance recommendation models, yet gradually causing their sizes exceed terabytes industrial recommender systems, and hence increase computing storage costs. To save resources while maintaining model performances, we propose SHARK, compression practice have summarized system scenarios. SHARK consists two main components. First, use novel first-order component Taylor expansion as importance scores prune number...
Abstract The cloud network has the advantages in efficiently offloading large‐scale Internet traffic, which is considered as a promising architecture to provide satisfactory multimedia services for mobile users. However, most current studies lack joint consideration of economic and security hybrid networks. In this paper, novel service optimization mechanism proposed hereby meet user's requirements mentioned above while guaranteeing reliability service. Firstly, credible scheme designed help...
Brain-inspired Spiking Neural Networks (SNNs) have the characteristics of event-driven and high energy-efficient, which are different from traditional Artificial (ANNs) when deployed on edge devices such as neuromorphic chips. Most previous work focuses SNNs training strategies to improve model performance brings larger deeper network architectures. It's difficult deploy these complex networks resource-limited directly. To meet demand, people compress very cautiously balance computation...
Large language models have demonstrated exceptional capability in natural understanding and generation. However, their generation speed is limited by the inherently sequential nature of decoding process, posing challenges for real-time applications. This paper introduces Lexical Unit Decoding (LUD), a novel methodology implemented data-driven manner, accelerating process without sacrificing output quality. The core our approach observation that pre-trained model can confidently predict...
Increasing the size of embedding layers has shown to be effective in improving performance recommendation models, yet gradually causing their sizes exceed terabytes industrial recommender systems, and hence increase computing storage costs. To save resources while maintaining model performances, we propose SHARK, compression practice have summarized system scenarios. SHARK consists two main components. First, use novel first-order component Taylor expansion as importance scores prune number...
The vBNS (very-high-performance Backbone Network Service) is a high-speed Internet Protocol (IP) over an ATM national backbone network. A "reserved bandwidth" service designed based on the network traffic characteristics as initial quality-of-service (QoS) offering for user community. This paper outlines improved configuration in layer and required new router mechanisms to support this service.