(1) Debiprasad Ghosh's answer to Why does batch normalization help? - Quora
When you have less than million parameters you can have luxury of global optimization or curvature-aided learning. But if you have more than million parameters, you have to work with gradient descent only. Whitening brings spherical shape to the error surface, where originally it is ellipsoidal and Gradient Descent works better if error surface is spherical than elliptical.
july 2017
« earlier      
3d admin agile ai ajax algorithm amazon-web-services analytics android angular.js animation api app apple appstore arduino artwork assembly automation backbone.js blog blogging boilerplate books business buttons c canvas catalog cheatsheet cli cloud cms cocoa coffeescript collaboration color comet command-line component components consulting cooking copywriting course crypto css css3 dashboard data date debugging deployment design devops diet dns docker dom ebook ec2 editor emacs email favorite-theme finance fonts food forms framework fun gamedev git github golang graphics grid hardware health hiring history hn hosting html html5 http humor icons images inspiration ios ios-design ipad iphone javascript jekyll jquery json keyboard languages layout learning library life linux livereload livereload-articles livereload-mentions livereload-tools logging lorem mac management markdown marketing math medicine meteor.js metro mobile mockko mockups mongodb monitoring mvc networking node.js nosql novice-designer-recommendations npm objective-c osx paas performance photoshop php physics placeholder plugin postgres pricing pride-and-prejudice procrastination productivity programming psychology python rails redis regexp ruby s3 saas sales scaling scraping screencast search security seo sharing shell ssh startup startup-essential statistics stock sublime-plugins sublime-text support swift sysops teaching template terminal testing theme tips tmux tool tools tutorial twitter twitter-bootstrap ui-design unix upload usability ux validation video vim visualization web webdev webkit windows writing wysiwyg x86 xcode

Copy this bookmark: