Commits

Breadcrumbs

History for

onmaster

User selector

All users

Datepicker

All time

Commit History

Commits on Aug 7, 2014

• cleaned up cmakelists.txt and prefix of lib

  Show description for a6d4e12
  kashif

  committed
  a6d4e12

  Copy full SHA for a6d4e12

Commits on May 27, 2014

• fixed shared lib. module name clash

  kashif

  committed
  6cc39b5

  Copy full SHA for 6cc39b5

Commits on Feb 16, 2014

• libname is convnet

  kashif

  committed
  66a7302

  Copy full SHA for 66a7302

Commits on Jan 26, 2014

• whitespace

  kashif

  committed
  504490c

  Copy full SHA for 504490c

Commits on Feb 3, 2012

• Convolution with minibatch size 32 or 64 is now about twice as fast. Added --conserve-mem parameter.

  Show description for f97e2a4
  

  akrizhevsky

  committed
  f97e2a4

  Copy full SHA for f97e2a4

Commits on Jan 11, 2012

• Image scaling with bilinear filtering.

  Show description for bcfc8e9
  

  akrizhevsky

  committed
  bcfc8e9

  Copy full SHA for bcfc8e9

Commits on Dec 21, 2011

• Included changes by Oriol Vinyals to allow compilation on Windows systems. Added elementwise max layer. Added linear/sqrt/square neurons.

  Show description for 7f28479
  

  akrizhevsky

  committed
  7f28479

  Copy full SHA for 7f28479

Commits on Dec 9, 2011

• Recurrent layer support. Sum layer support. All layers with weights take multiple inputs. Decoupled neuron activation functions from fc/conv/local layers.

  Show description for 8520655
  

  akrizhevsky

  committed
  8520655

  Copy full SHA for 8520655

Commits on Oct 15, 2011

• Fixed recently-introduced crashing bug on save. Added ability to unshare convolutional units.

  Show description for c54fce3
  

  akrizhevsky

  committed
  c54fce3

  Copy full SHA for c54fce3

Commits on Oct 6, 2011

• shownet now only takes its own options

  Show description for fbaa0d7
  

  akrizhevsky@gmail.com

  committed
  fbaa0d7

  Copy full SHA for fbaa0d7

Commits on Oct 4, 2011

• --show-preds now takes softmax layer name itself, so it's no longer necessary to supply --logreg-name

  Show description for ea5bae3
  

  akrizhevsky@gmail.com

  committed
  ea5bae3

  Copy full SHA for ea5bae3

Commits on Oct 3, 2011

• Rewrote shownet so that it extends the model class now, making it easy to run the model from that script. Added ability to show predictions made by the net.

  Show description for 9e5e5b4
  

  akrizhevsky@gmail.com

  committed
  9e5e5b4

  Copy full SHA for 9e5e5b4

Commits on Sep 7, 2011

• Code organization

  Show description for 43de8c1
  

  akrizhevsky@gmail.com

  committed
  43de8c1

  Copy full SHA for 43de8c1

Commits on Sep 1, 2011

• Clarified what the data provider is doing a bit. Also renamed convdp.py to convdata.py to be consistent with data.py

  Show description for 2ce23f6
  

  akrizhevsky@gmail.com

  committed
  2ce23f6

  Copy full SHA for 2ce23f6

Commits on Aug 30, 2011

• Do some error checking on data type returned by data provider

  Show description for 7947a8d
  

  akrizhevsky@gmail.com

  committed
  7947a8d

  Copy full SHA for 7947a8d

• Linesep screwup

  Show description for 87f41cb
  

  akrizhevsky@gmail.com

  committed
  87f41cb

  Copy full SHA for 87f41cb

Commits on Jul 23, 2011

• BSD license

  Show description for 8b7b1c3
  

  akrizhevsky@gmail.com

  committed
  8b7b1c3

  Copy full SHA for 8b7b1c3

Commits on Jul 19, 2011

• License

  Show description for c0a9f04
  

  akrizhevsky@gmail.com

  committed
  c0a9f04

  Copy full SHA for c0a9f04

Commits on Jul 7, 2011

• cleaned up bprop a bit. now the individual layers don't have to explicitly call bprop on the layers below them.

  Show description for 418bd76
  

  akrizhevsky@gmail.com

  committed
  418bd76

  Copy full SHA for 418bd76

Commits on Jul 6, 2011

• got rid of mfi for good. also added data provider that trains on translations of cifar10

  Show description for de9f700
  

  akrizhevsky@gmail.com

  committed
  de9f700

  Copy full SHA for de9f700

• removed restriction that minibatch size must be a multiple of 128. also removed mfi output order from cudaconv2, since it was unused.

  Show description for 9137e5a
  

  akrizhevsky@gmail.com

  committed
  9137e5a

  Copy full SHA for 9137e5a

Commits on Jul 5, 2011

• introduced worker classes so convnet's main loop isn't a jumble of if statements. also templated Data, so now it's CPUData and GPUData

  Show description for 7244a1a
  

  akrizhevsky@gmail.com

  committed
  7244a1a

  Copy full SHA for 7244a1a

Commits on Jul 1, 2011

• delete gc files from trunk, they're downloads now

  Show description for 7f2a1c7
  

  akrizhevsky@gmail.com

  committed
  7f2a1c7

  Copy full SHA for 7f2a1c7

• nicer output

  Show description for 0db702c
  

  akrizhevsky@gmail.com

  committed
  0db702c

  Copy full SHA for 0db702c

Commits on Jun 30, 2011

• added readme

  Show description for 288a861
  

  akrizhevsky@gmail.com

  committed
  288a861

  Copy full SHA for 288a861

• default param files

  Show description for 83430d9
  

  akrizhevsky@gmail.com

  committed
  83430d9

  Copy full SHA for 83430d9

• git-svn-id: http://cuda-convnet.googlecode.com/svn/trunk@27 bc73d74b-de6f-cad7-86a4-6ddd3d5d919c

  

  akrizhevsky@gmail.com

  committed
  522676a

  Copy full SHA for 522676a

• git-svn-id: http://cuda-convnet.googlecode.com/svn/trunk@13 bc73d74b-de6f-cad7-86a4-6ddd3d5d919c

  

  akrizhevsky@gmail.com

  committed
  173d333

  Copy full SHA for 173d333

Commits on Jun 29, 2011

• initial commit

  Show description for 90753db
  

  akrizhevsky@gmail.com

  committed
  90753db

  Copy full SHA for 90753db

Loading