| .. Licensed to the Apache Software Foundation (ASF) under one |
| or more contributor license agreements. See the NOTICE file |
| distributed with this work for additional information |
| regarding copyright ownership. The ASF licenses this file |
| to you under the Apache License, Version 2.0 (the |
| "License"); you may not use this file except in compliance |
| with the License. You may obtain a copy of the License at |
| |
| http://www.apache.org/licenses/LICENSE-2.0 |
| |
| Unless required by applicable law or agreed to in writing, |
| software distributed under the License is distributed on an |
| "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY |
| KIND, either express or implied. See the License for the |
| specific language governing permissions and limitations |
| under the License. |
| |
| |
| 开发时间表 |
| ==================== |
| |
| .. csv-table:: |
| :header: "版本","模块","特性" |
| |
| "v0.1 2015 九月 ","神经网络 ","前向传播神经网络, 包括 CNN, MLP" |
| " "," ","类 RBM 模型, 包括 RBM" |
| " "," ","循环神经网络, 包括标准 RNN" |
| " ","架构 ","在单节点运行一个工作组 (包括划分)" |
| " "," ","在单节点运行多个工作组, 用 `Hogwild <http://www.eecs.berkeley.edu/~brecht/papers/hogwildTR.pdf>`_ " |
| " "," ","分布式 Hogwild" |
| " "," ","跨多节点运行多个工作组 , 如 `Downpour <http://papers.nips.cc/paper/4687-large-scale-distritbuted-deep-networks>`_" |
| " "," ","All-Reduce 训练架构如 `DeepImage <http://arxiv.org/abs/1501.02876>`_ " |
| " "," ","服务器间负载均衡" |
| " ","失败恢复 ","检查点和恢复" |
| " ","工具 ","用 GNU 自动工具安装" |
| "v0.2 2016 一月 ","神经网络 ","前向传播神经网络, 包括 AlexNet, cuDNN 层, 工具" |
| " "," ","循环神经网络, 包括 GRU 层和 BPTT" |
| " "," ","模型划分和混合划分" |
| " ","工具 ","融合 Mesos 资源管理" |
| " "," ","准备部署 Docker images" |
| " "," ","可视化神经网络和调试信息" |
| " ","绑定 ","主要组件 Python 绑定" |
| " ","GPU ","单节点多个 GPU " |
| "v0.3 2016 四月 ","GPU ","多个节点, 每个包含多个 GPU" |
| " "," ","用 GPU 和 CPU 混合训练 `CcT <http://arxiv.org/abs/1504.04343>`_" |
| " "," ","支持 cuDNN v4 " |
| " ","安装 ","删除 ZeroMQ, CZMQ 依赖, 单节点训练 zookeeper" |
| " ","优化器 ","添加新的 SGD 优化器,包括 Adam, AdamMax 和 AdaDelta" |
| " ","绑定 ","增强 Python 绑定训练" |
| "v1.0 2016 九月 ","模型抽象 ","Tensor 基于线性代数, 神经网络和随机运算" |
| " "," ","分布式参数更新优化器" |
| " ","硬件 ","使用 Cuda 和 Cudnn for Nvidia GPU" |
| " "," ","使用 OpenCL for AMD GPU 及其他设备" |
| " ","跨平台 ","从 Linux 扩展到 MacOS" |
| " "," ","大型图像模型, 例如, `VGG <https://arxiv.org/pdf/1409.1556.pdf>`_ 和 `Residual Net <http://arxiv.org/abs/1512.03385>`_" |
| "v1.1 2017 一月 ","模型库 ","GoogleNet; 医疗健康模型" |
| " ","Caffe 转换器 ","使用 SINGA 训练模型, 从 caffe proto 文件配置" |
| " ","模型组件 ","添加 concat 和 slice 层; 接受多个输入网络" |
| " ","编译和安装 ","Windows 支持" |
| " "," ","通过与 protobuf 和 openblas 一起编译 SINGA 简化安装" |
| " "," ","用 Jenkins 自动生成 python wheel" |
| " "," ","从 Debian packages 安装 SINGA" |
| "v1.2 2018 六月 ","AutoGrad ","后向传播 AutoGrad" |
| " ","Python 3 ","PySinga 支持 Python 3" |
| " ","模型 ","添加流行模型, 包括 VGG, ResNet, DenseNet, InceptionNet" |