Commits

Breadcrumbs

History for

onmaster

User selector

All users

All time

Commit History

Commits on Jan 26, 2016

• Merge pull request #3132 from bwilbertz/LastIterationLoss

  Show description for f5e28fe
  jeffdonahue

  committed
  f5e28fe

  Copy full SHA for f5e28fe

Commits on Nov 26, 2015

• replace snprintf with a C++98 equivalent

  eelstork

  committed
  b72b031

  Copy full SHA for b72b031

Commits on Nov 12, 2015

• Fix loss of last iteration when average_loss > 1

  Show description for dc48870
  bwilbertz

  committed
  dc48870

  Copy full SHA for dc48870

Commits on Oct 17, 2015

• Add automatic upgrade for solver type

  ronghanghu

  committed
  c1f7fe1

  Copy full SHA for c1f7fe1

• Split solver code into one file per solver class

  ronghanghu

  committed
  b822a70

  Copy full SHA for b822a70

Commits on Sep 18, 2015

• refine format of switch case in solver

  yanchen036

  committed
  f75d594

  Copy full SHA for f75d594

Commits on Sep 14, 2015

• Merge pull request #3049 from seanbell/check-snapshot-prefix

  Show description for e4baef2
  ronghanghu

  committed
  e4baef2

  Copy full SHA for e4baef2

• Check that the snapshot directory is writeable before starting training

  

  Sean Bell

  committed
  ab554cb

  Copy full SHA for ab554cb

Commits on Sep 9, 2015

• Minor: missing space in string formatting

  

  Sean Bell

  committed
  8fbac04

  Copy full SHA for 8fbac04

Commits on Aug 22, 2015

• Add signal handler and early exit/snapshot to Solver.

  Show description for ff19d5f
  jyegerlehner

  committed
  ff19d5f

  Copy full SHA for ff19d5f

Commits on Aug 14, 2015

• Adam solver

  Show description for 4e4c89b
  

  PatWie

  authored and

  ronghanghu

  committed
  4e4c89b

  Copy full SHA for 4e4c89b

Commits on Aug 13, 2015

• Merge pull request #2903 from ronghanghu/multi_gpu

  Show description for bb0a90e
  ronghanghu

  committed
  bb0a90e

  Copy full SHA for bb0a90e

Commits on Aug 12, 2015

• Data Layers Parallel for Multi-GPU

  Show description for 0d34d5b
  ronghanghu

  committed
  0d34d5b

  Copy full SHA for 0d34d5b

Commits on Aug 10, 2015

• Clean up and modernize AdaDelta code; add learning rate support; add additional test cases

  matthiasplappert

  committed
  f2e523e

  Copy full SHA for f2e523e

• Updated AdaDelta for modern Caffe; reduced iterations on multi-iter tests

  

  kevinbache

  authored and

  matthiasplappert

  committed
  4c58741

  Copy full SHA for 4c58741

• Implement AdaDelta; add test cases; add mnist examples

  

  mohomran

  authored and

  matthiasplappert

  committed
  1ce3380

  Copy full SHA for 1ce3380

Commits on Aug 9, 2015

• Multi-GPU

  Show description for e5575cf
  

  cypof

  authored and

  shelhamer

  committed
  e5575cf

  Copy full SHA for e5575cf

• Use net_->learnable_params() instead of net_->params() in RMSprop

  Show description for b97ca6d
  ronghanghu

  committed
  b97ca6d

  Copy full SHA for b97ca6d

• Implement RMSProp Solver

  Show description for abe99e8
  

  erogol

  authored and

  ronghanghu

  committed
  abe99e8

  Copy full SHA for abe99e8

Commits on Aug 7, 2015

• Net: add learnable_params_ used by solvers to correctly handle shared params

  Show description for d5b42bf
  jeffdonahue

  committed
  d5b42bf

  Copy full SHA for d5b42bf

• Snapshot model weights/solver state to HDF5 files.

  Show description for ada055b
  erictzeng

  committed
  ada055b

  Copy full SHA for ada055b

Commits on May 28, 2015

• directly normalize accumulated gradients

  Show description for 0e7a078
  shelhamer

  committed
  0e7a078

  Copy full SHA for 0e7a078

Commits on May 27, 2015

• adjust local learning rate and decay according to gradient accumulation

  Show description for 55585f5
  shelhamer

  committed
  55585f5

  Copy full SHA for 55585f5

• zero-init param diffs in gradient checker

  

  longjon

  authored and

  shelhamer

  committed
  539f879

  Copy full SHA for 539f879

• zero-init param diffs and accumulate gradients

  Show description for 41cf06c
  

  longjon

  authored and

  shelhamer

  committed
  41cf06c

  Copy full SHA for 41cf06c

• Solver::MakeUpdate() -> Solver::ApplyUpdate

  Show description for 76db47e
  shelhamer

  committed
  76db47e

  Copy full SHA for 76db47e

• deduplicate decay and local rate in solver updates

  shelhamer

  committed
  a85f7f1

  Copy full SHA for a85f7f1

• Refactor solvers regularization and logging code

  

  cypof

  authored and

  shelhamer

  committed
  ca81667

  Copy full SHA for ca81667

Commits on May 14, 2015

• Merge pull request #2072 from jeffdonahue/final-snapshot-off-by-one

  Show description for 07f56a0
  longjon

  committed
  07f56a0

  Copy full SHA for 07f56a0

Commits on Mar 13, 2015

• Fix for solver issue pointed out by @moskewcz in #1972

  Show description for e46cf85
  jeffdonahue

  committed
  e46cf85

  Copy full SHA for e46cf85

Commits on Mar 9, 2015

• Increment iter_ before snapshotting, remove +1 logic -- fixes final

  Show description for 4a4118f
  jeffdonahue

  committed
  4a4118f

  Copy full SHA for 4a4118f

Commits on Mar 3, 2015

• Blobs are ND arrays (for N not necessarily equals 4).

  Show description for 1434e87
  jeffdonahue

  committed
  1434e87

  Copy full SHA for 1434e87

Commits on Feb 17, 2015

• no phase for the solver to orchestrate

  shelhamer

  committed
  ab0cc51

  Copy full SHA for ab0cc51

Commits on Feb 14, 2015

• Add gradient clipping -- limit L2 norm of parameter gradients

  jeffdonahue

  committed
  f38ddef

  Copy full SHA for f38ddef

Commits on Jan 30, 2015

• Enhancements for debug_info to display more information.

  Show description for 2a43133
  jeffdonahue

  committed
  2a43133

  Copy full SHA for 2a43133