• The principle of logarithmic probability regression and its implementation in Python


    catalog 1、 Logarithmic probability and logarithmic probability regression 2、 Sigmoid function 3、 Maximum likelihood method 4、 Gradient descent method 4、 Python implementation 1、 Logarithmic probability and logarithmic probability regression    in logarithmic probability regression, we output the model of the sample\(y^*\)It is defined that the sample is a positive exampleprobability, will\(\frac{y^*}{1-y^*}\)Defined asprobability(odds)The probability is the […]

  • Saving deep learning: a deep learning method with insufficient labeled data


    Abstract:To solve the problem of data dependence in deep learning and reduce the cost of data annotation has become a research hotspot in the industry. This paper will introduce the following research directions: semi supervised / weakly supervised learning, data synthesis, active learning and self supervision. 1. Introduction Thanks to the development of deep learning, […]

  • Why are residual networks effective and what are their developments?


    Why are residual networks effective and what are their developments? -Amaze2’s answer – Zhihu https://www.zhihu.com/questio… Gao Piao’s answer has clearly introduced the residual network. Here is a new improvement of the residual network, that isResidual shrinkage network。 1. Motivation (redundant information everywhere) When carrying out machine learning tasks, our dataset often contains some redundant information […]

  • Practice of machine learning based on Flink in microblog


    About Weibo Weibo, launched in 2008, is the mainstream social media platform in China, with 222 million daily users and 516 million monthly users, providing users with services of online creation, sharing and discovery of high-quality content; at present, the large-scale machine learning platform of Weibo can support hundreds of billions of parameters and millions […]

  • MNIST dataset download and visualization


    Introduction to MNIST dataset Official website of MNIST dataset: http://yann.lecun.com/exdb/mnist/ MNIST database is a very classic data set, just like you learn to write a “Hello word” program at the beginning of programming, learn deep learning, you will write a model to recognize MNIST data set. MNIST data set is composed of 0-9 handwritten digital […]

  • Introduction to Gan


    Introduction to Gan Author: Mei Haoming 1. Principle introduction The full name of Gan is generative adversarial network, that is, generative adversarial network. Generative learning a generative model; adversal uses confrontation training; networks uses neural networks. Gan model is a kind ofLearning data distribution through confrontationOfGenerative modelIts core idea is to pass theGenerative network G(generator) […]

  • Service Custom index in kubernetes


    Prometheus indicator type Counter CounterType represents an indicator of monotonically increasing sample data, i.e. only increase but not decrease, unless the monitoring system is reset. For example, promql can be used to analyze the change rate of the main function generated by HTTP. Get the growth rate of HTTP requests rate(http_requests_total[5m]) HTTP request address of […]

  • Sequential anomaly detection of ant group intelligent monitoring: anomaly detection based on CNN neural network


    1、 Background In the field of ant group intelligent monitoring, sequential anomaly detection is a very important link. In the implementation of anomaly detection, the business side outputs metrics index data according to industry standards, and monitors various indicators of different businesses, applications, interfaces and clusters, including metrics indicators (total amount, failure amount, time consumption, […]

  • Linear discriminant analysis, LDA


    Linear discriminant classifier consists of vector $W $and deviation term $B $. Given the example $x $, it predicts the category tag $y $according to the following rules, that is$y=sign(w^Tx+b)$The column vector is represented by lowercase and row vector is represented by transposition.The classification process is divided into two steps Firstly, the weight vector w […]

  • How to generate online machine learning samples based on Flink?


    Compared with offline machine learning, online machine learning has better performance in the timeliness of model updating, model iteration cycle, business experiment effect and so on. Therefore, it has become an effective way to improve business indicators to migrate machine learning from offline to online. In online machine learning, sample is the key link. This […]

  • [deep learning] understand the deep residual contraction network in 10 minutes


    RESNET, the residual network, won the best paper award of IEEE Conference on computer vision and pattern recognition in 2016, and has been cited 38295 times in Google. Deep residual shrinkage network is an improved version of deep residual network, which is actually the integration of deep residual network, attention mechanism and soft threshold function. […]

  • Activation function of attention mechanism: adaptive parameterized relu activation function


    On the basis of summarizing the traditional activation function and attention mechanism, this paper interprets an activation function under attention mechanism, namely adaptive parametric corrector linear unit (aprelu). 1. Activation function Activation function is an important part of artificial neural network, its function is to realize the nonlinearity of artificial neural network. We first introduce […]