RELEASE_NOTES

Release Notes - SINGA - Version singa-4.1.0

SINGA is a distributed deep learning library.

This release includes following changes:

  * New examples
    * Add an example for malaria detection using cell images.
    * Add an example for structured data learning.

  * Add support for models running on top of RDBMS
    * Add support for in-database model definition and selection in RDBMS.
    * Implement training-free model evaluation metrics for in-database model selection.
    * Implement a coordinator to balance between training-free and training-based model evaluations 
      for in-database model selection.
    
  * Enhance distributed training
    * Add implementations for the sum error loss.
    * Improve the optimizer to return model gradients.
    * Improve the iterative checking for tensors and strings in the ModelMeta class.
    
  * Enhance example code
    * Add support for flexible setting of training configurations for models, e.g., learning rates, 
      weight decay, momentum, etc.
    * Add implementations for dynamic models with varying layer sizes.

  * Update the website
    * Add illustrations for database integration.
    * Update users of Apache SINGA.

  * Fix bugs
    * Update the NVIDIA_GPGKEY in the Dockerfile for building wheel files.
    * Update the versions of dependencies in the wheel file.
    * Fix the collections module in the model.py file.

----------------------------------------------------------------------------------------------

Release Notes - SINGA - Version singa-4.0.0

SINGA is a distributed deep learning library.

This release includes following changes:

  * Enhance distributed training
    * Add support for configuration of number of GPUs to be used.
    * Increase max epoch for better convergence.
    * Print intermediate mini-batch information.
    * Add support for switching between CPU and GPU devices.

  * Enhance example code
    * Update the args of normalize forward function in the transforms of the BloodMnist example.
    * Update the xceptionnet in the cnn example.
    * Add arguments for weight decay, momentum and learning rates in the cnn example.
    * Add training scripts for more datasets and model types in the cnn example.
    * Add resnet dist version for the large dataset cnn example.
    * Add cifar 10 multi process for the large dataset cnn example.
    * Add sparsification implementation for mnist in the large dataset cnn example.
    * Update the cifar datasets downloading to local directories.
    * Extend the cifar datasets load function for customized directorires.

  * Enhance the webpage
    * Update online documentation for distributed training.

  * Promote code quality
    * Update inline comments for prepreocessing and data loading

  * Update the PIL image module

  * Update the runtime Dockerfile

  * Update the conda files

----------------------------------------------------------------------------------------------

Release Notes - SINGA - Version singa-3.3.0

SINGA is a distributed deep learning library.

This release includes following changes:

  * New examples:
    * Add one CNN example for the BloodMnist dataset, a sub set of MedMNIST.
    * Add one example for the medical image analysis.

  * Enhance distributed training
    * Add key information printing, e.g., throughput and communication time, for distributed training.
    * Optimize printing and logging modules for faster distributed training.

  * Enhance example code
    * Add more datasets and model implementations for the cifar_distributed_cnn example.
    * Update the running script for the cifar_distributed_cnn example to include more models.
    * Update the dataset path for the largedataset_cnn example for more flexibility.
    * Add more model implementations for the largedataset_cnn example.

  * Enhance the webpage
    * Reconstruct the singa webpage to include project features.
    * Update the Git web site by deploying it via .asf.yaml.
    * Update the Chinese and Vietnamese documentations.

  * Debug and add assertions for input tensor data types in the opt.py.

  * Change pointer type to void for generalizing data types.

  * Fix bugs
    * Fix the python test error due to operations not implemented for some data types.
    * Fix the model of pad from bytes to str.

----------------------------------------------------------------------------------------------

Release Notes - SINGA - Version singa-3.2.0

SINGA is a distributed deep learning library.

This release includes following changes:

  * New examples:
    * Add one cifar-10 distributed CNN example for benchmarking the performance of the distributed
      training.
    * Add one large CNN example for training with a dataset from the filesysetm.

  * Enhance distributed training
    * Improve the data augmentation module for faster distributed training.
    * Add device synchronization for more accurate time measurements during the distributed training.

  * Add Support for half-precision floating-point format (fp16) in deep learning models and 
    computational kernels.

  * Update new onnx APIs and fix onnx examples accordingly, namely, DenseNet121, ShuffleNetv1, 
    ShuffleNetv2, SqueezeNet, VGG19.

  * Add a new method to resize images by given width and height.

  * Use docusaurus versioning to simplify the process of generating the project homepage.

  * Promote code quality
    * Unify the formats of docstrings that describe the contents and usage of the module.
    * Unify the parameters of command-line arguments.

  * Fix bugs
    * Fix the CI build error by downloading the tbb binaries.
    * Add disabling graph option for accessing parameter or gradient tensors during distributed  
      training.
    * Solve the warnings of deprecated functions in the distributed optimizer module.

----------------------------------------------------------------------------------------------

Release Notes - SINGA - Version singa-3.1.0

SINGA is a distributed deep learning library.

This release includes following changes:

  * Tensor core:
    * Support tensor transformation (reshape, transpose) for tensors up to 6 dimensions.
    * Implement traverse_unary_transform in Cuda backend, which is similar to CPP backend one.

  * Add new tensor operators into the autograd module, including
    CosSim, DepthToSpace, Embedding, Erf, Expand, Floor, Pad, Round, Rounde, SpaceToDepth, UpSample, Where. 
    The corresponding ONNX operators are thus supported by SINGA.

  * Add Embedding and Gemm into the layer module.

  * Add SGD operators to opt module, including RMSProp, Adam, and AdaGrad.

  * Extend the sonnx module to support 
    DenseNet121, ShuffleNetv1, ShuffleNetv2, SqueezeNet, VGG19, GPT2, and RoBERTa,

  * Reconstruct sonnx to 
    * Support creating operators from both layer and autograd.
    * Re-write SingaRep to provide a more powerful intermediate representation of SINGA.
    * Add a SONNXModel which implements from Model to provide uniform API and features.

  * Add one example that trains a BiLSTM model over the InsuranceQA data.

  * Replace the Travis CI with Github workflow. Add quality and coverage management.

  * Add compiling and packaging scripts to creat wheel packages for distribution.

  * Fix bugs
    * Fix IMDB LSTM model example training script.
    * Fix Tensor operation Mult on Broadcasting use cases.
    * Gaussian function on Tensor now can run on Tensor with odd size.
    * Updated a testing helper function gradients() in autograd to lookup param gradient by param python object id for testing purpose.


----------------------------------------------------------------------------------------------

Release Notes - SINGA - Version singa-3.0.0

SINGA is a distributed deep learning library.

This release includes following changes:

  * Code quality has been promoted by introducing linting check in CI and auto code formatter. 
    For linting, the tools, `cpplint` and `pylint`, are used and configured to comply 
    [google coding styles](http://google.github.io/styleguide/)  details in `tool/linting/`. 
    Similarly, formatting tools, `clang-format` and `yapf` configured with google coding styles, 
    are the recommended one for developers to clean code before submitting changes, 
    details in `tool/code-format/`. [LGTM](https://lgtm.com) is enabled on Github for 
    code quality check; License check is also enabled.

 * New Tensor APIs are added for naming consistency, and feature enhancement: 
   - size(), mem_size(), get_value(), to_proto(), l1(), l2(): added for the sake of naming consistency
   - AsType(): convert data type between `float` and `int`
   - ceil(): perform element-wise ceiling of the input
   - concat(): concatenate two tensor
   - index selector: e.g. tensor1[:,:,1:,1:]
   - softmax(in, axis): allow to perform softmax on a axis on a multi-dimensional tensor

  * 14 new operators are added into the autograd module: Gemm, GlobalAveragePool, ConstantOfShape, 
    Dropout, ReduceSum, ReduceMean, Slice, Ceil, Split, Gather, Tile, NonZero, Cast, OneHot. 
    Their unit tests are added as well.

  * 14 new operators are added to sonnx module for both backend and frontend: 
    [Gemm](https://github.com/onnx/onnx/blob/master/docs/Operators.md#Gemm), 
    [GlobalAveragePool](https://github.com/onnx/onnx/blob/master/docs/Operators.md#GlobalAveragePool), 
    [ConstantOfShape](https://github.com/onnx/onnx/blob/master/docs/Operators.md#ConstantOfShape), 
    [Dropout](https://github.com/onnx/onnx/blob/master/docs/Operators.md#Dropout), 
    [ReduceSum](https://github.com/onnx/onnx/blob/master/docs/Operators.md#ReduceSum), 
    [ReduceMean](https://github.com/onnx/onnx/blob/master/docs/Operators.md#ReduceMean), 
    [Slice](https://github.com/onnx/onnx/blob/master/docs/Operators.md#Slice), 
    [Ceil](https://github.com/onnx/onnx/blob/master/docs/Operators.md#Ceil), 
    [Split](https://github.com/onnx/onnx/blob/master/docs/Operators.md#Split), 
    [Gather](https://github.com/onnx/onnx/blob/master/docs/Operators.md#Gather), 
    [Tile](https://github.com/onnx/onnx/blob/master/docs/Operators.md#Tile), 
    [NonZero](https://github.com/onnx/onnx/blob/master/docs/Operators.md#NonZero), 
    [Cast](https://github.com/onnx/onnx/blob/master/docs/Operators.md#Cast), 
    [OneHot](https://github.com/onnx/onnx/blob/master/docs/Operators.md#OneHot). 
    Their tests are added as well.

  * Some ONNX models are imported into SINGA, including 
    [Bert-squad](https://github.com/onnx/models/tree/master/text/machine_comprehension/bert-squad), 
    [Arcface](https://github.com/onnx/models/tree/master/vision/body_analysis/arcface), 
    [FER+ Emotion](https://github.com/onnx/models/tree/master/vision/body_analysis/emotion_ferplus), 
    [MobileNet](https://github.com/onnx/models/tree/master/vision/classification/mobilenet), 
    [ResNet18](https://github.com/onnx/models/tree/master/vision/classification/resnet), 
    [Tiny Yolov2](https://github.com/onnx/models/tree/master/vision/object_detection_segmentation/tiny_yolov2), 
    [Vgg16](https://github.com/onnx/models/tree/master/vision/classification/vgg), and Mnist.

  * Some operators now support [multidirectional broadcasting](https://github.com/onnx/onnx/blob/master/docs/Broadcasting.md#multidirectional-broadcasting), 
    including Add, Sub, Mul, Div, Pow, PRelu, Gemm 

  * [Distributed training with communication optimization]. [DistOpt](./python/singa/opt.py) 
    has implemented multiple optimization techniques, including gradient sparsification, 
    chunk transmission, and gradient compression.

  * Computational graph construction at the CPP level. The operations submitted to the Device are buffered.
    After analyzing the dependency, the computational graph is created, which is further analyzed for
    speed and memory optimization. To enable this feature, use the [Module API](./python/singa/module.py).

  * New website based on Docusaurus. The documentation files are moved to a separate repo [singa-doc](https://github.com/apache/singa-doc).
    The static website files are stored at [singa-site](https://github.com/apache/singa-site).

  * DNNL([Deep Neural Network Library](https://github.com/intel/mkl-dnn)), powered by Intel, 
    is integrated into `model/operations/[batchnorm|pooling|convolution]`, 
    the changes is opaque to the end users. The current version is dnnl v1.1 
    which replaced previous integration of mkl-dnn v0.18. The framework could 
    boost the performance of dl operations when executing on CPU. The dnnl dependency 
    is installed through conda.

  * Some Tensor APIs are marked as deprecated which could be replaced by broadcast, 
    and it can support better on multi-dimensional operations. These APIs are
    add_column(), add_row(), div_column(), div_row(), mult_column(), mult_row()

  * Conv and Pooling are enhanced to support fine-grained padding like (2,3,2,3), 
    and [SAME_UPPER, SAME_LOWER](https://github.com/onnx/onnx/blob/master/docs/Operators.md#Conv) 
    pad mode and shape checking.

  * Reconstruct soonx, 
    - Support two types of weight value (Initializer and Constant Node); 
    - For some operators (BatchNorm, Reshape, Clip, Slice, Gather, Tile, OneHot), 
      move some inputs to its attributes; 
    - Define and implement the type conversion map. 

------------------------------------------------------------------------
Release Notes - SINGA - Version singa-incubating-2.0.0

SINGA is a general distributed deep learning platform for training big deep
learning models over large datasets.

This release includes following features:

  * Core components
    * [SINGA-434] Support tensor broadcasting
    * [SINGA-370] Improvement to tensor reshape and various misc. changes related to SINGA-341 and 351

  * Model components
    * [SINGA-333] Add support for Open Neural Network Exchange (ONNX) format
    * [SINGA-385] Add new python module for optimizers
    * [SINGA-394] Improve the CPP operations via Intel MKL DNN lib
    * [SINGA-425] Add 3 operators , Abs(), Exp() and leakyrelu(), for Autograd 
    * [SINGA-410] Add two function, set_params() and get_params(), for Autograd Layer class
    * [SINGA-383] Add Separable Convolution for autograd
    * [SINGA-388] Develop some RNN layers by calling tiny operations like matmul, addbias.
    * [SINGA-382] Implement concat operation for autograd    
    * [SINGA-378] Implement maxpooling operation and its related functions for autograd
    * [SINGA-379] Implement batchnorm operation and its related functions for autograd

  * Utility functions and CI
    * [SINGA-432] Update depdent lib versions in conda-build config
    * [SINGA-429] Update docker images for latest cuda and cudnn
    * [SINGA-428] Move Docker images under Apache user name

  * Documentation and usability
    * [SINGA-395] Add documentation for autograd APIs
    * [SINGA-344] Add a GAN example
    * [SINGA-390] Update installation.md
    * [SINGA-384] Implement ResNet using autograd API
    * [SINGA-352] Complete SINGA documentation in Chinese version
      
  * Bugs fixed
    * [SINGA-431] Unit Test failed - Tensor Transpose
    * [SINGA-422] ModuleNotFoundError: No module named "_singa_wrap"
    * [SINGA-418] Unsupportive type 'long' in python3.  
    * [SINGA-409] Basic `singa-cpu` import throws error
    * [SINGA-408] Unsupportive function definition in python3
    * [SINGA-380] Fix bugs from Reshape  

---------------------------------------------------------------
Release Notes - SINGA - Version singa-incubating-1.2.0

SINGA is a general distributed deep learning platform for training big deep
learning models over large datasets.

This release includes following features:

  * Core components
      * [SINGA-290] Upgrade to Python 3
      * [SINGA-341] Added stride functionality to tensors for CPP
      * [SINGA-347] Create a function that supports einsum
      * [SINGA-351] Added stride support and cudnn codes to cuda

  * Model components
      * [SINGA-300] Add residual networks for imagenet classification
      * [SINGA-312] Rename layer parameters
      * [SINGA-313] Add L2 norm layer
      * [SINGA-315] Reduce memory footprint by Python generator for parameter
      * [SINGA-316] Add SigmoidCrossEntropy
      * [SINGA-324] Extend RNN layer to accept variant seq length across batches
      * [SINGA-326] Add Inception V4 for ImageNet classification
      * [SINGA-328] Add VGG models for ImageNet classification
      * [SINGA-329] Support layer freezing during training (fine-tuning)
      * [SINGA-346] Update cudnn from V5 to V7
      * [SINGA-349] Create layer operations for autograd
      * [SINGA-363] Add DenseNet for Imagenet classification

  * Utility functions and CI
      * [SINGA-274] Improve Debian packaging with CPack
      * [SINGA-303] Create conda packages
      * [SINGA-337] Add test cases for code
      * [SINGA-348] Support autograd MLP Example
      * [SINGA-345] Update Jenkins and fix bugs in compliation
      * [SINGA-354] Update travis scripts to use conda-build for all platforms
      * [SINGA-358] Consolidated RUN steps and cleaned caches in Docker containers
      * [SINGA-359] Create alias for conda packages

  * Documentation and usability
      * [SINGA-223] Fix side navigation menu in the website
      * [SINGA-294] Add instructions to run CUDA unit tests on Windows
      * [SINGA-305] Add jupyter notebooks for SINGA V1 tutorial
      * [SINGA-319] Fix link errors on the index page
      * [SINGA-352] Complete SINGA documentation in Chinese version
      * [SINGA-361] Add git instructions for contributors and committers

  * Bugs fixed
      * [SINGA-330] fix openblas building on i7 7700k
      * [SINGA-331] Fix the bug of tensor division operation
      * [SINGA-350] Error from python3 test
      * [SINGA-356] Error using travis tool to build SINGA on mac os
      * [SINGA-363] Fix some bugs in imagenet examples
      * [SINGA-368] Fix the bug in Cifar10 examples
      * [SINGA-369] the errors of examples in testing

---------------------------------------------------------------
Release Notes - SINGA - Version singa-incubating-1.1.0

SINGA is a general distributed deep learning platform for training big deep learning models over large datasets.

This release includes following features:

  * Core components
      * [SINGA-296] Add sign and to_host function for pysinga tensor module

  * Model components
      * [SINGA-254] Implement Adam for V1
      * [SINGA-264] Extend the FeedForwardNet to accept multiple inputs
      * [SINGA-267] Add spatial mode in batch normalization layer
      * [SINGA-271] Add Concat and Slice layers
      * [SINGA-275] Add Cross Entropy Loss for multiple labels
      * [SINGA-278] Convert trained caffe parameters to singa
      * [SINGA-287] Add memory size check for cudnn convolution

  * Utility functions and CI
      * [SINGA-242] Compile all source files into a single library.
      * [SINGA-244] Separating swig interface and python binding files
      * [SINGA-246] Imgtool for image augmentation
      * [SINGA-247] Add windows support for singa
      * [SINGA-251] Implement image loader for pysinga
      * [SINGA-252] Use the snapshot methods to dump and load models for pysinga
      * [SINGA-255] Compile mandatory depedent libaries together with SINGA code
      * [SINGA-259] Add maven pom file for building java classes
      * [SINGA-261] Add version ID into the checkpoint files
      * [SINGA-266] Add Rafiki python toolkits
      * [SINGA-273] Improve license and contributions
      * [SINGA-284] Add python unittest into Jenkins and link static libs into whl file
      * [SINGA-280] Jenkins CI support
      * [SINGA-288] Publish wheel of PySINGA generated by Jenkins to public servers

  * Documentation and usability
      * [SINGA-263] Create Amazon Machine Image
      * [SINGA-268] Add IPython notebooks to the documentation
      * [SINGA-276] Create docker images
      * [SINGA-289] Update SINGA website automatically using Jenkins
      * [SINGA-295] Add an example of image classification using GoogleNet

  * Bugs fixed
      * [SINGA-245] float as the first operand can not multiply with a tensor object
      * [SINGA-293] Bug from compiling PySINGA on Mac OS X with multiple version of Python

---------------------------------------------------------------
Release Notes - SINGA - Version singa-incubating-1.0.0

SINGA is a general distributed deep learning platform for training big deep learning models over large datasets.

This release includes following features:

  * Core abstractions including Tensor and Device
      * [SINGA-207]  Update Tensor functions for matrices
      * [SINGA-205]  Enable slice and concatenate operations for Tensor objects
      * [SINGA-197]  Add CNMem as a submodule in lib/
      * [SINGA-196]  Rename class Blob to Block
      * [SINGA-194]  Add a Platform singleton
      * [SINGA-175]  Add memory management APIs and implement a subclass using CNMeM
      * [SINGA-173]  OpenCL Implementation
      * [SINGA-171]  Create CppDevice and CudaDevice
      * [SINGA-168]  Implement Cpp Math functions APIs
      * [SINGA-162]  Overview of features for V1.x
      * [SINGA-165]  Add cross-platform timer API to singa
      * [SINGA-167]  Add Tensor Math function APIs
      * [SINGA-166]  light built-in logging for making glog optional
      * [SINGA-164]  Add the base Tensor class


  * IO components for file read/write, network and data pre-processing
      * [SINGA-233]  New communication interface
      * [SINGA-215]  Implement Image Transformation for Image Pre-processing
      * [SINGA-214]  Add LMDBReader and LMDBWriter for LMDB
      * [SINGA-213]  Implement Encoder and Decoder for CSV
      * [SINGA-211]  Add TextFileReader and TextFileWriter for CSV files
      * [SINGA-210]  Enable checkpoint and resume for v1.0
      * [SINGA-208]  Add DataIter base class and a simple implementation
      * [SINGA-203]  Add OpenCV detection for cmake compilation
      * [SINGA-202]  Add reader and writer for binary file
      * [SINGA-200]  Implement Encoder and Decoder for data pre-processing



  * Module components including layer classes, training algorithms and Python binding
      * [SINGA-235]  Unify the engines for cudnn and singa layers
      * [SINGA-230]  OpenCL Convolution layer and Pooling layer
      * [SINGA-222]  Fixed bugs in IO
      * [SINGA-218]  Implementation for RNN CUDNN version
      * [SINGA-204]  Support the training of feed-forward neural nets
      * [SINGA-199]  Implement Python classes for SGD optimizers
      * [SINGA-198]  Change Layer::Setup API to include input Tensor shapes
      * [SINGA-193]  Add Python layers
      * [SINGA-192]  Implement optimization algorithms for SINGA v1 (nesterove, adagrad, rmsprop)
      * [SINGA-191]  Add "autotune" for CudnnConvolution Layer
      * [SINGA-190]  Add prelu layer and flatten layer
      * [SINGA-189]  Generate python outputs of proto files
      * [SINGA-188]  Add Dense layer
      * [SINGA-187]  Add popular parameter initialization methods
      * [SINGA-186]  Create Python Tensor class
      * [SINGA-184]  Add Cross Entropy loss computation
      * [SINGA-183]  Add the base classes for optimizer, constraint and regularizer
      * [SINGA-180]  Add Activation layer and Softmax layer
      * [SINGA-178]  Add Convolution layer and Pooling layer
      * [SINGA-176]  Add loss and metric base classes
      * [SINGA-174]  Add Batch Normalization layer and Local Response Nomalization layer.
      * [SINGA-170]  Add Dropout layer and CudnnDropout layer.
      * [SINGA-169]  Add base Layer class for V1.0


  * Examples
      * [SINGA-232]  Alexnet on Imagenet
      * [SINGA-231]  Batchnormlized VGG model for cifar-10
      * [SINGA-228]  Add Cpp Version of Convolution and Pooling layer
      * [SINGA-227]  Add Split and Merge Layer and add ResNet Implementation

  * Documentation
      * [SINGA-239]  Transfer documentation files of v0.3.0 to github
      * [SINGA-238]  RBM on mnist
      * [SINGA-225]  Documentation for installation and Cifar10 example
      * [SINGA-223]  Use Sphinx to create the website

  * Tools for compilation and some utility code
      * [SINGA-229]  Complete install targets
      * [SINGA-221]  Support for Travis-CI
      * [SINGA-217]  build python package with setup.py
      * [SINGA-216]  add jenkins for CI support
      * [SINGA-212]  Disable the compilation of libcnmem if USE_CUDA is OFF
      * [SINGA-195]  Channel for sending training statistics
      * [SINGA-185]  Add CBLAS and GLOG detection for singav1
      * [SINGA-181]  Add NVCC supporting for .cu files
      * [SINGA-177]  Add fully cmake supporting for the compilation of singa_v1
      * [SINGA-172]  Add CMake supporting for Cuda and Cudnn libs

----------------------------------------------------------
Release Notes - SINGA - Version singa-incubating-0.3.0

SINGA is a general distributed deep learning platform for training big deep learning models over large datasets.

This release includes following features:

  * GPU Support
    * [SINGA-131] Implement and optimize hybrid training using both CPU and GPU
    * [SINGA-136] Support cuDNN v4
    * [SINGA-134] Extend SINGA to run over a GPU cluster
    * [SINGA-157] Change the priority of cudnn library and install libsingagpu.so

  * Remove Dependencies
    * [SINGA-156] Remove the dependency on ZMQ for single process training
    * [SINGA-155] Remove zookeeper for single-process training

  * Python Binding
    * [SINGA-126] Python Binding for Interactive Training

  * Other Improvements
    * [SINGA-80] New Blob Level and Address Level Math Operation Interface
    * [SINGA-130] Data Prefetching
    * [SINGA-145] New SGD based optimization Updaters: AdaDelta, Adam, AdamMax

  * Bugs Fixed
    * [SINGA-148] Race condition between Worker threads and Driver
    * [SINGA-150] Mesos Docker container failed
    * [SIGNA-141] Undesired Hash collision when locating process id to worker…
    * [SINGA-149] Docker build fail
    * [SINGA-143] The compilation cannot detect libsingagpu.so file


-----------------------------------------
Release Notes - SINGA - Version singa-incubating-0.2.0

SINGA is a general distributed deep learning platform for training big deep learning models over large datasets. It is
designed with an intuitive programming model based on the layer abstraction. SINGA supports a wide variety of popular
deep learning models.

This release includes following features:

  * Programming model
    * [SINGA-80] New Blob Level and Address Level Math Operation Interface
    * [SINGA-82] Refactor input layers using data store abstraction
    * [SINGA-87] Replace exclude field to include field for layer configuration
    * [SINGA-110] Add Layer member datavec_ and gradvec_
    * [SINGA-120] Implemented GRU and BPTT (BPTTWorker)

  * Neuralnet layers
    * [SINGA-91] Add SoftmaxLayer and ArgSortLayer
    * [SINGA-106] Add dummy layer for test purpose
    * [SINGA-120] Implemented GRU and BPTT (GRULayer and OneHotLayer)

  * GPU training support
    * [SINGA-100] Implement layers using CUDNN for GPU training
    * [SINGA-104] Add Context Class
    * [SINGA-105] Update GUN make files for compiling cuda related code
    * [SINGA-98] Add Support for AlexNet ImageNet Classification Model


  * Model/Hybrid partition
    * [SINGA-109] Refine bridge layers
    * [SINGA-111] Add slice, concate and split layers
    * [SINGA-113] Model/Hybrid Partition Support

  * Python binding
    * [SINGA-108] Add Python wrapper to singa

  * Predict-only mode
    * [SINGA-85] Add functions for extracting features and test new data

  * Integrate with third-party tools
    * [SINGA-11] Start SINGA on Apache Mesos
    * [SINGA-78] Use Doxygen to generate documentation
    * [SINGA-89] Add Docker support

  * Unit test
    * [SINGA-95] Add make test after building

  * Other improvment
    * [SINGA-84] Header Files Rearrange
    * [SINGA-93] Remove the asterisk in the log tcp://169.254.12.152:*:49152
    * [SINGA-94] Move call to google::InitGoogleLogging() from Driver::Init() to main()
    * [SINGA-96] Add Momentum to Cifar10 Example
    * [SINGA-101] Add ll (ls -l) command in .bashrc file when using docker
    * [SINGA-114] Remove short logs in tmp directory
    * [SINGA-115] Print layer debug information in the neural net graph file
    * [SINGA-118] Make protobuf LayerType field id easy to assign
    * [SIGNA-97] Add HDFS Store

  * Bugs fixed
    * [SINGA-85] Fix compilation errors in examples
    * [SINGA-90] Miscellaneous trivial bug fixes
    * [SINGA-107] Error from loading pre-trained params for training stacked RBMs
    * [SINGA-116] Fix a bug in InnerProductLayer caused by weight matrix sharing

-----------------------------------------

Features included in singa-incubating-0.1.0:

  * Job management
    * [SINGA-3] Use Zookeeper to check stopping (finish) time of the system
    * [SINGA-16] Runtime Process id Management
    * [SINGA-25] Setup glog output path
    * [SINGA-26] Run distributed training in a single command
    * [SINGA-30] Enhance easy-to-use feature and support concurrent jobs
    * [SINGA-33] Automatically launch a number of processes in the cluster
    * [SINGA-34] Support external zookeeper service
    * [SINGA-38] Support concurrent jobs
    * [SINGA-39] Avoid ssh in scripts for single node environment
    * [SINGA-43] Remove Job-related output from workspace
    * [SINGA-56] No automatic launching of zookeeper service
    * [SINGA-73] Refine the selection of available hosts from host list

  * Installation with GNU Auto tool
    * [SINGA-4] Refine thirdparty-dependency installation
    * [SINGA-13] Separate intermediate files of compilation from source files
    * [SINGA-17] Add root permission within thirdparty/install.
    * [SINGA-27] Generate python modules for proto objects
    * [SINGA-53] Add lmdb compiling options
    * [SINGA-62] Remove building scrips and auxiliary files
    * [SINGA-67] Add singatest into build targets

  * Distributed training
    * [SINGA-7] Implement shared memory Hogwild algorithm
    * [SINGA-8] Implement distributed Hogwild
    * [SINGA-19] Slice large Param objects for load-balance
    * [SINGA-29] Update NeuralNet class to enable layer partition type customization
    * [SINGA-24] Implement Downpour training framework
    * [SINGA-32] Implement AllReduce training framework
    * [SINGA-57] Improve Distributed Hogwild

  * Training algorithms for different model categories
    * [SINGA-9] Add Support for Restricted Boltzman Machine (RBM) model
    * [SINGA-10] Add Support for Recurrent Neural Networks (RNN)

  * Checkpoint and restore
    * [SINGA-12] Support Checkpoint and Restore

  * Unit test
    * [SINGA-64] Add the test module for utils/common

  * Programming model
    * [SINGA-36] Refactor job configuration, driver program and scripts
    * [SINGA-37] Enable users to set parameter sharing in model configuration
    * [SINGA-54] Refactor job configuration to move fields in ModelProto out
    * [SINGA-55] Refactor main.cc and singa.h
    * [SINGA-61] Support user defined classes
    * [SINGA-65] Add an example of writing user-defined layers

  * Other features
    * [SINGA-6] Implement thread-safe singleton
    * [SINGA-18] Update API for displaying performance metric
    * [SINGA-77] Integrate with Apache RAT

Some bugs are fixed during the development of this release
  * [SINGA-2] Check failed: zsock_connect
  * [SINGA-5] Server early terminate when zookeeper singa folder is not initially empty
  * [SINGA-15] Fixg a bug from ConnectStub function which gets stuck for connecting layer_dealer_
  * [SINGA-22] Cannot find openblas library when it is installed in default path
  * [SINGA-23] Libtool version mismatch error.
  * [SINGA-28] Fix a bug from topology sort of Graph
  * [SINGA-42] Issue when loading checkpoints
  * [SINGA-44] A bug when reseting metric values
  * [SINGA-46] Fix a bug in updater.cc to scale the gradients
  * [SINGA-47] Fix a bug in data layers that leads to out-of-memory when group size is too large
  * [SINGA-48] Fix a bug in trainer.cc that assigns the same NeuralNet instance to workers from diff groups
  * [SINGA-49] Fix a bug in HandlePutMsg func that sets param fields to invalid values
  * [SINGA-66] Fix bugs in Worker::RunOneBatch function and ClusterProto
  * [SINGA-79] Fix bug in singatool that can not parse -conf flag

Features planned for the next release
  * [SINGA-11] Start SINGA using Mesos
  * [SINGA-31] Extend Blob to support xpu (cpu or gpu)
  * [SINGA-35] Add random number generators
  * [SINGA-40] Support sparse Param update
  * [SINGA-41] Support single node single GPU training