Commits

Branch selector

 2c7997f

User selector

All users

Datepicker

All time

Commit History

Commits on Jan 28, 2014

• fix a few issues with dropout

  liuliu

  committed
  2c7997f

  Copy full SHA for 2c7997f

Commits on Jan 27, 2014

• choose a more reasonable learn_rate

  liuliu

  committed
  0399811

  Copy full SHA for 0399811

• I think that I can run image-net now

  Show description for 9068159
  liuliu

  committed
  9068159

  Copy full SHA for 9068159

Commits on Jan 25, 2014

• generating test result in on-going basis

  liuliu

  committed
  9b42e74

  Copy full SHA for 9b42e74

Commits on Jan 24, 2014

• default to initialization on w, bias and wnum

  liuliu

  committed
  852ace3

  Copy full SHA for 852ace3

• verified write / read works

  liuliu

  committed
  517b56f

  Copy full SHA for 517b56f

Commits on Jan 23, 2014

• compile zero dependency sqlite3

  liuliu

  committed
  82411fc

  Copy full SHA for 82411fc

• get sqlite3 storage works for ccv

  liuliu

  committed
  1eab672

  Copy full SHA for 1eab672

• fix variance computing issue within ccv_dpm

  liuliu

  committed
  1055b87

  Copy full SHA for 1055b87

Commits on Jan 22, 2014

• add sqlite3 to ccv, I am still shuffle the fact that if you includes

  Show description for 519365e
  liuliu

  committed
  519365e

  Copy full SHA for 519365e

• clean up how we free memory for convnet

  liuliu

  committed
  6b542f2

  Copy full SHA for 6b542f2

• done right on actually free resources

  liuliu

  committed
  e46a13a

  Copy full SHA for e46a13a

Commits on Jan 18, 2014

• change name from batch_unit to a more sensible unit

  liuliu

  committed
  d4b9643

  Copy full SHA for d4b9643

• change batch_per_block from 4 to 8

  liuliu

  committed
  639ee92

  Copy full SHA for 639ee92

• about as fast as Alex on CIFAR

  Show description for e7ddd47
  liuliu

  committed
  e7ddd47

  Copy full SHA for e7ddd47

Commits on Jan 17, 2014

• still slower than Alex on CIFAR,

  Show description for 900be7c
  liuliu

  committed
  900be7c

  Copy full SHA for 900be7c

• verified my output

  liuliu

  committed
  598a238

  Copy full SHA for 598a238

Commits on Jan 16, 2014

• using scratch space for multi way weight updates as well

  liuliu

  committed
  fc74d4a

  Copy full SHA for fc74d4a

• using scratch space

  liuliu

  committed
  290de1b

  Copy full SHA for 290de1b

• about 7.8ms per image, need to verify the output though

  liuliu

  committed
  3bc49a4

  Copy full SHA for 3bc49a4

• optimized weight update a bit, need to verify (again)

  liuliu

  committed
  6afcc1c

  Copy full SHA for 6afcc1c

Commits on Jan 15, 2014

• seems this way to bprop weight is faster, but need to verify this actually is correct

  liuliu

  committed
  0d5213c

  Copy full SHA for 0d5213c

• remove input_loads non-sense

  liuliu

  committed
  16c4478

  Copy full SHA for 16c4478

• try to optimize image-net performance

  liuliu

  committed
  d95dfb7

  Copy full SHA for d95dfb7

Commits on Jan 13, 2014

• updated with verify

  liuliu

  committed
  47225de

  Copy full SHA for 47225de

• I believe this is correct dropout impl, ready to impl image-net

  liuliu

  committed
  98f4c0e

  Copy full SHA for 98f4c0e

Commits on Jan 12, 2014

• added THANKS

  liuliu

  committed
  9ab4dd0

  Copy full SHA for 9ab4dd0

• refactored a bit to put more stuff in cwc_convnet_context_t

  liuliu

  committed
  ab46a73

  Copy full SHA for ab46a73

Commits on Jan 11, 2014

• have rnorm checked and implemented on GPU

  liuliu

  committed
  4688c9a

  Copy full SHA for 4688c9a

Commits on Jan 10, 2014

• added a quick CPU impl on rnorm, will do the GPU impl once I verified the CPU correctness with CIFAR-10 data

  liuliu

  committed
  b773b56

  Copy full SHA for b773b56

• beautify the output

  liuliu

  committed
  514914a

  Copy full SHA for 514914a

• about 0.5 slower

  liuliu

  committed
  24a0e2e

  Copy full SHA for 24a0e2e

• average pooling back prop claimed more shared memory than it should

  liuliu

  committed
  a2519c1

  Copy full SHA for a2519c1

• I managed to get performance closer to Alex one (about 0.8 slower)

  liuliu

  committed
  fabcaab

  Copy full SHA for fabcaab

Commits on Jan 9, 2014

• the GPU version runs

  liuliu

  committed
  964492b

  Copy full SHA for 964492b

Pagination

Previous

Next