A5102796060- Stochastic Gradient Optimization Techniques
- Sparse and Compressive Sensing Techniques
- Machine Learning and ELM
- Domain Adaptation and Few-Shot Learning
- Advanced Neural Network Applications
- Anomaly Detection Techniques and Applications
- Reinforcement Learning in Robotics
- Markov Chains and Monte Carlo Methods
- Advanced Bandit Algorithms Research
- Face and Expression Recognition
- Optimization and Search Problems
- Neural Networks and Applications
- Distributed systems and fault tolerance
- Advanced Image Processing Techniques
- Distributed and Parallel Computing Systems
- Interconnection Networks and Systems
- Cloud Computing and Resource Management
- Software System Performance and Reliability
- COVID-19 diagnosis using AI
- Machine Learning and Algorithms
- Generative Adversarial Networks and Image Synthesis
- Gaussian Processes and Bayesian Inference
- Aerospace Engineering and Control Systems
- Wireless Power Transfer Systems
- Network Security and Intrusion Detection
Shenzhen Institutes of Advanced Technology
2025
University of Chinese Academy of Sciences
2025
Guangdong Provincial Center for Disease Control and Prevention
2025
Simons Foundation
2024
University of California, Berkeley
2024
Johns Hopkins University
2019-2021
Peking University
2018-2019
King University
2019
Dalian University
2016
East China Normal University
2006
Programmable packet scheduling enables algorithms to be programmed into the data plane without changing hardware. Existing proposals either have no hardware implementations for switch ASICs or require multiple strict-priority queues.
ABSTRACT Background Cloud‐native applications are increasingly becoming popular in modern software design. Employing a microservice‐based architecture into these is prevalent strategy that enhances system availability and flexibility. However, cloud‐native introduce new challenges, including frequent inter‐service communication the management of heterogeneous codebases hardware, resulting unpredictable complexity dynamism. Furthermore, as scale, only limited research teams or enterprises...
Understanding the behavior of stochastic gradient descent (SGD) in context deep neural networks has raised lots concerns recently. Along this line, we study a general form based optimization dynamics with unbiased noise, which unifies SGD and standard Langevin dynamics. Through investigating dynamics, analyze on escaping from minima its regularization effects. A novel indicator is derived to characterize efficiency through measuring alignment noise covariance curvature loss function. Based...
Insulator is an extremely important component of the power transmission system. This article adopts model convolutional neural network from recent studies on deep learning to achieve end-to-end intelligent detection insulators, which helps computers identify insulator footage faster and obtain fault more accurate insulator. Firstly, method object used determine location insulator; then extrapolated using fully networks; lastly, based insulator's explosion characteristic, coordinates can be...
<title>Abstract</title> Background. Our study aimed to evaluate vaccine effectiveness (VE) of a live oral pentavalent rotavirus (RotaTeq, RV5) among < 5-year-old children in three provinces China via propensity score matched test negative design case-control study. Methods. We used electronic health records and person-matched immunization information system data obtain on acute gastroenteritis (AGE) cases Guangdong, Beijing, Hubei evaluated for AGE emergency department settings tested...
Compared with standard supervised learning, the key difficulty in semi-supervised learning is how to make full use of unlabeled data. A recently proposed method, virtual adversarial training (VAT), smartly performs without label information impose a local smoothness on classifier, which especially beneficial learning. In this work, we propose tangent-normal regularization (TNAR) as an extension VAT by taking data manifold into consideration. The TNAR composed two complementary parts, tangent...
The gradient noise of SGD is considered to play a central role in the observed strong generalization abilities deep learning. While past studies confirm that magnitude and covariance structure are critical for regularization, it remains unclear whether or not class distributions important. In this work we provide negative results by showing noises classes different from can also effectively regularize descent. Our finding based on novel observation noise: multiplication matrix sampling...
Abstract In order to improve the linkage of related auxiliary systems in intelligent substation and realize complete monitoring control functions equipment station, this paper integrates technologies Internet Things, edge computing big data establish an system IoT substation. The overall structure is introduced detail. effective calculation reduction task a single node achieved by application function algorithm balancing allocation scheduling, effectively improving immediacy security Things....
There is an increasing realization that algorithmic inductive biases are central in preventing overfitting; empirically, we often see a benign overfitting phenomenon overparameterized settings for natural learning algorithms, such as stochastic gradient descent (SGD), where little to no explicit regularization has been employed. This work considers this issue arguably the most basic setting: constant-stepsize SGD (with iterate averaging or tail averaging) linear regression regime. Our main...
In this paper we consider multi-objective reinforcement learning where the objectives are balanced using preferences. practice, preferences often given in an adversarial manner, e.g., customers can be picky many applications. We formalize problem as episodic on a Markov decision process, transitions unknown and reward function is inner product of preference vector with pre-specified functions. two settings. online setting, agent receives (adversarial) every episode proposes policies to...
Understanding the algorithmic bias of \emph{stochastic gradient descent} (SGD) is one key challenges in modern machine learning and deep theory. Most existing works, however, focus on \emph{very small or even infinitesimal} rate regime, fail to cover practical scenarios where \emph{moderate annealing}. In this paper, we make an initial attempt characterize particular regularization effect SGD moderate regime by studying its behavior for optimizing overparameterized linear regression problem....
The prediction model of existing human body composition based on measured bioelectricity has problems that include redundant influence factors and low accuracy. To address these problems, this paper put forward a Akaike Information Criterion (AIC) improved entropy method. First, combining with the AIC information principle, we selected set characteristic parameters from physiological arguments, constructed model; Second, method was used to solve unknown coefficients in predictive model, then...
For the problem of task-agnostic reinforcement learning (RL), an agent first collects samples from unknown environment without supervision reward signals, then is revealed with a and asked to compute corresponding near-optimal policy. Existing approaches mainly concern worst-case scenarios, in which no structural information reward/transition-dynamics utilized. Therefore best sample upper bound $\propto\widetilde{\mathcal{O}}(1/\epsilon^2)$, where $\epsilon>0$ target accuracy obtained...
Empirically, large-scale deep learning models often satisfy a neural scaling law: the test error of trained model improves polynomially as size and data grow. However, conventional wisdom suggests consists approximation, bias, variance errors, where increases with size. This disagrees general form laws, which predict that increasing monotonically performance. We study theory laws in an infinite dimensional linear regression setup. Specifically, we consider $M$ parameters function sketched...
WITHDRAWAL STATEMENT bioRxiv has withdrawn this manuscript because it inappropriately listed Peter L. Bartlett, Jingfeng Wu, and Bin Yu as authors. Therefore, work should not be cited reference for the project. If you have any questions, please contact corresponding author.
Cloud-native applications are increasingly becoming popular in modern software design. Employing a microservice-based architecture into these is prevalent strategy that enhances system availability and flexibility. However, cloud-native also introduce new challenges, such as frequent inter-service communication the complexity of managing heterogeneous codebases hardware, resulting unpredictable dynamism. Furthermore, scale, only limited research teams or enterprises possess resources for...