Commits

Branch selector

 master

User selector

All users

Datepicker

All time

Commit History

Commits on Sep 24, 2014

• Merge pull request karpathy#15 from rylans/master

  Show description for 3eba1b9
  karpathy

  committed
  3eba1b9

  Copy full SHA for 3eba1b9

• Fix small typo

  rylans

  committed
  566e584

  Copy full SHA for 566e584

Commits on Sep 16, 2014

• fixing small bug, removing the restriction to have a full batch finished before we can predict

  karpathy

  committed
  ba84e46

  Copy full SHA for ba84e46

• tweaking Readme substantially

  karpathy

  committed
  62c0a04

  Copy full SHA for 62c0a04

• some refactoring, and allow regression layer to take a number in backprop because this is a common mistake otherwise

  karpathy

  committed
  7786e7c

  Copy full SHA for 7786e7c

• Net fixes and small cleanup

  karpathy

  committed
  ec183b1

  Copy full SHA for ec183b1

• tweaking input layer to have sensible defaults

  karpathy

  committed
  ea9d6a9

  Copy full SHA for ea9d6a9

• removing legacy test. switching to jasmine

  karpathy

  committed
  fcf0621

  Copy full SHA for fcf0621

Commits on Sep 1, 2014

• updating the readme

  karpathy

  committed
  1e32d97

  Copy full SHA for 1e32d97

• adding a small speedtest

  karpathy

  committed
  735db5d

  Copy full SHA for 735db5d

• 2x convlayer optimizations, thanks to @mdda

  karpathy

  committed
  f8c8d4f

  Copy full SHA for f8c8d4f

• updaing cifar10 demo to be much nicer

  karpathy

  committed
  aa7fdba

  Copy full SHA for aa7fdba

Commits on Aug 31, 2014

• Merge branch 'master' of https://github.com/karpathy/convnetjs

  karpathy

  committed
  f2e7e29

  Copy full SHA for f2e7e29

• fixing the build script and tweaking mnist demo tiny bit

  karpathy

  committed
  a5d0fd4

  Copy full SHA for a5d0fd4

• removing compiled files from repo, these will be found in releases from now on

  karpathy

  committed
  c366cbe

  Copy full SHA for c366cbe

Commits on Jun 30, 2014

• Merge pull request karpathy#8 from levicc00123/master

  Show description for 34eef10
  karpathy

  committed
  34eef10

  Copy full SHA for 34eef10

Commits on Jun 28, 2014

• Added a JSON file that allows this package to be added to the bower package manager.

  levicc00123

  committed
  f2cb00e

  Copy full SHA for f2cb00e

Commits on May 16, 2014

• adding some testing code. we use jasmie since it looks nice. This commit only includes a basic gradcheck but I hope to build on this foundation for more extensive tests of all library features in t…

  Show description for f1addd3
  karpathy

  committed
  f1addd3

  Copy full SHA for f1addd3

Commits on Apr 26, 2014

• Extending Vol() constructor to be able to take a list as convenience.

  karpathy

  committed
  5284e9b

  Copy full SHA for 5284e9b

• Adding MagicNet demo files

  karpathy

  committed
  a75797f

  Copy full SHA for a75797f

• haha oops forgot to include the new MagicNet class

  karpathy

  committed
  74d2e3f

  Copy full SHA for 74d2e3f

• Adding MagicNet for fully automatic blackbox learning: performs n-fold cross-validation over training data for many randomly sampled neural networks and forms a prediction ensemble from the best pe…

  Show description for 3495bca
  karpathy

  committed
  3495bca

  Copy full SHA for 3495bca

Commits on Apr 13, 2014

• Refreshing all demos to the most recent and current state

  

  karpathy

  committed
  f50101b

  Copy full SHA for f50101b

• removing learning rate as a modifier for Adadelta to be consistent with the paper

  

  karpathy

  committed
  2258e38

  Copy full SHA for 2258e38

Commits on Apr 10, 2014

• Adding tanh non-linearity

  

  karpathy

  committed
  6a2a69f

  Copy full SHA for 6a2a69f

Commits on Apr 1, 2014

• adding an SVM classifier layer, can be used instead of softmax

  

  karpathy

  committed
  3c6b7e8

  Copy full SHA for 3c6b7e8

Commits on Mar 31, 2014

• fixing an issue with loading nets from json: they used to incorrectly initialize with non-typed javascript arrays resulting in decreased performance. This fix basically makes loaded nets from JSON …

  Show description for 2b65387
  

  karpathy

  committed
  2b65387

  Copy full SHA for 2b65387

• Adding Adagrad and Adadelta as ways of training the networks, and a demo that compares all the different approaches on MNIST dataset. This also required a new MultiGraph in build/vis, which I rathe…

  Show description for 73ae935
  

  karpathy

  committed
  73ae935

  Copy full SHA for 73ae935

Commits on Mar 30, 2014

• adding a quadratic input transformation layer, which essentially allows us to train tensor layers. Essentially, ever input x is extended to also contain all the binary interaction terms x_ix_j. Thi…

  Show description for b25a012
  

  karpathy

  committed
  b25a012

  Copy full SHA for b25a012

Commits on Mar 29, 2014

• Adding image regression demo and tweaking the docs

  

  karpathy

  committed
  8a759b1

  Copy full SHA for 8a759b1

• Refreshing the 2D demo to be nicer and the network to be changeable.

  

  karpathy

  committed
  6309cca

  Copy full SHA for 6309cca

• adding maxout activation neurons. They take a set of input neurons and max over groups of them. In a convnet, this happens only along depth preserving spatial dimensions x,y. The group size can be …

  Show description for 8aea15c
  

  karpathy

  committed
  8aea15c

  Copy full SHA for 8aea15c

• fixing small introduced bugs, and a bit of documentation

  

  karpathy

  committed
  3a833c8

  Copy full SHA for 3a833c8

• Changing the way convolution/pooling layers work. They now do VALID convolution by default, and a new option to add padding is included. This is not a backwards compatible change unfortunatelly, an…

  Show description for 19bcc3c
  

  karpathy

  committed
  19bcc3c

  Copy full SHA for 19bcc3c

Commits on Feb 23, 2014

• fixing a stride issue with convolutional layer

  

  karpathy

  committed
  a369c4a

  Copy full SHA for a369c4a

Pagination

Previous

Next