Curriculum learning (CL) is a training strategy that trains machine model from easier data to harder data, which imitates the meaningful order in human curricula. As an easy-to-use plug-in, CL has demonstrated its power improving generalization capacity and convergence rate of various models wide range scenarios such as computer vision natural language processing etc. In this survey article, we comprehensively review aspects including motivations, definitions, theories, applications. We...

The gradient descent algorithm is a type of optimization that widely used to solve machine learning model parameters. Through continuous iteration, it obtains the objective function, gradually approaches optimal solution and finally minimum loss function related frequently in process logical regression, which common binary classification approach. This paper compares analyzes differences between batch its derivative algorithms — stochastic mini- terms iteration number, through experiments,...

Deep neural networks are known to be data-driven and label noise can have a marked impact on model performance. Recent studies shown great robustness classic image recognition even under high noisy rate. In medical applications, learning from datasets with is more challenging since imaging tend asymmetric (class-dependent) suffer observer variability. this paper, we systematically discuss define the two common types of in images - disagreement inconsistency expert opinions single-target...

Curriculum learning (CL) is a machine paradigm gradually from easy to hard, which inspired by human curricula. As an easy-to-use and general training strategy, CL has been widely applied various multimedia tasks covering images, texts, audios, videos, etc. The effectiveness of recently facilitated increasing number new algorithms. However, there no open-source library for curriculum learning, making it hard reproduce, evaluate compare the numerous algorithms on fair benchmarks settings. To...

For the multisensor linear discrete time‐invariant stochastic control systems with different measurement matrices and correlated noises, centralized fusion white noise estimators are presented by minimum variance criterion under condition that input matrix is full column rank. They have expensive computing burden due to high‐dimension extended matrix. To reduce burden, weighted presented. It proved same accuracy as estimators, so it has global optimality. can be applied signal processing in...

Under the background of Energy Internet, ever-growing scale electric power system has brought new challenges and opportunities. Numerous categories measurement data, as cornerstone communication, play a crucial role in security stability system. However, present sampling transmission equipment inevitably suffers from data missing, which seriously degrades stable operation state estimation. Therefore, this paper, we consider load an example first develop missing detection algorithm terms...

In the framework of evidence theory, one open and crucial issues is how to determine basic probability assignment (BPA), which directly related whether decision result correct. This paper proposes a novel method for obtaining BPA based on Adaboost. The uses training data generate multiple strong classifiers each attribute model, used singleton proposition since weights classification provide necessary information fundamental hypotheses. composite quantified by calculating area ratio...

Abstract This paper improves the Dynamic Time Warping (DTW) algorithm. In order to change problem of traditional search range DTW speech recognition algorithm is too large, an improved proposed limit path. Firstly, analyzed find its path, distortion and efficiency. Secondly, introduced, path limited, simulation carried out in MATLAB. Compare improvement results with before comparing efficiency, efficiency calculated. Experimental show that new superior overall performance.

For the linear discrete stochastic systems with multiple sensors and unknown noise statistics, an online estimators of variances cross‐covariances are designed by using measurement feedback, full‐rank decomposition, weighted least squares theory. Further, a self‐tuning fusion Kalman filter is presented. The Fadeeva formula used to establish ARMA innovation model statistics. sampling correlated function stationary reversible identify It proved that presented converges optimal filter, which...

With the development and improvement of hydraulic steering system, articulated system became research focus numerous domestic foreign scholars. The full with a compact s... | Find, read cite all you need on Tech Science Press

In the application of Dempster-Shafer (D-S) evidence theory, determination basic probability assignment (BPA) is a key step. How to determine BPA an open issue. To solve this problem, new method based on distribution proposed in paper. First, approximate training set constructed. Next, distance according distributions set. Then similarity degree calculated distance. Finally, normalized obtain BPA. The effectiveness our shown by classifying dataset Iris.

This article has been retracted. A retraction notice can be found at https://doi.org/10.3233/JIFS-219433.

Abstract Field Programmable Gate Array (FPGA) devices are widely used in the field of Digital Signal Processing (DSP), and algorithm DSP has been many fields such as speech signal processing, audio image information system, control system so on. In development design modern digital systems, use programmable logic is becoming more common. There ways to implement processing based on FPGA: first one with VHDL or Verilog HDL language; second mature IP core encapsulated by FPGA company; third EDA...

In this paper, a cable aging state assessment method based on the combination of XGBoost and FocalLoss was proposed. Firstly, function is used to deal with problem small sample data serious imbalance in ratio between classes. Secondly, as custom loss algorithm, two can achieve effective status by extracting key features cables. Finally, it verified example that evaluation effectively number samples imbalance, accuracy significantly improved, which provide new direction for evaluation.

Aiming to the problem that traditional Dempster-Shafer(D-S) evidence theory may produce contradiction with intuition when synthesizing large conflict. A new combination method of conflict based on AHP and TOPSIS is proposed. Firstly, idea AHP, decision maker determines judgment matrix according relative importance each by combining subjective experience, weight vector obtained using AHP. Secondly, comprehensive evaluation index, i.e. TOPSIS. Finally, original data weighted modified fused...