v0.1 Sep 2015 |
Neural Network |
Feed forward neural network, including CNN, MLP |
|
|
RBM-like model, including RBM |
|
|
Recurrent neural network, including standard RNN |
|
Architecture |
One worker group on single node (with data partition) |
|
|
Multi worker groups on single node using Hogwild |
|
|
Distributed Hogwild |
|
|
Multi groups across nodes, like Downpour |
|
|
All-Reduce training architecture like DeepImage |
|
|
Load-balance among servers |
|
Failure recovery |
Checkpoint and restore |
|
Tools |
Installation with GNU auto Tools |
v0.2 Jan 2016 |
Neural Network |
Feed forward neural network, including AlexNet, cuDNN layers,Tools |
|
|
Recurrent neural network, including GRULayer and BPTT |
|
|
Model partition and hybrid partition |
|
Tools |
Integration with Mesos for resource management |
|
|
Prepare Docker images for deployment |
|
|
Visualization of neural net and debug information |
|
Binding |
Python binding for major components |
|
GPU |
Single node with multiple GPUs |
v0.3 April 2016 |
GPU |
Multiple nodes, each with multiple GPUs |
|
|
Heterogeneous training using both GPU and CPU CcT |
|
|
Support cuDNN v4 |
|
Installation |
Remove dependency on ZeroMQ, CZMQ, Zookeeper for single node training |
|
Updater |
Add new SGD updaters including Adam, AdamMax and AdaDelta |
|
Binding |
Enhance Python binding for training |
v1.0 Sep 2016 |
Programming abstraction |
Tensor with linear algebra, neural net and random operations |
|
|
Updater for distributed parameter updating |
|
Hardware |
Use Cuda and Cudnn for Nvidia GPU |
|
|
Use OpenCL for AMD GPU or other devices |
|
Cross-platform |
To extend from Linux to MacOS |
|
|
Large image models, e.g., VGG and Residual Net |
v1.1 Jan 2017 |
Model Zoo |
GoogleNet; Health-care models |
|
Caffe converter |
Use SINGA to train models configured in caffe proto files |
|
Model components |
Add concat and slice layers; accept multiple inputs to the net |
|
Compilation and installation |
Windows suppport |
|
|
Simplify the installation by compiling protobuf and openblas together with SINGA |
|
|
Build python wheel automatically using Jenkins |
|
|
Install SINGA from Debian packages |
v1.2 June 2018 |
AutoGrad |
AutoGrad for BP |
|
Python 3 |
Support Python 3 for PySinga |
|
Models |
Add popular models, including VGG, ResNet, DenseNet, InceptionNet |