• 欢迎访问速搜资源吧,如果在网站上找不到你需要的资源,可以在留言板上留言,管理员会尽量满足你!

【速搜问答】Chainer是什么

问答 admin 3年前 (2020-08-12) 822次浏览 已收录 0个评论

汉英对照:
Chinese-English Translation:

Chainer是一个开源的深度学习框架,完全在NumPy和CuPy Python库的基础上用Python编写。该开发工作由日本风险公司Preferred Networks与IBM,英特尔,微软和Nvidia合作进行。 Chainer 是一个开源的深度学习框架,完全在 NumPy 和 CuPy Python 库的基础上用 Python 编写。该开发工作由日本风险公司 Preferred Networks 与 IBM,英特尔,微软和 Nvidia 合作进行。

Chainer is an open source deep learning framework, written in Python based on numpy and cupy Python libraries. The development was conducted by preferred networks, a Japanese venture company, in partnership with IBM, Intel, Microsoft and NVIDIA. Chainer is an open source deep learning framework, written in Python based on numpy and cupy Python libraries. The development was conducted by preferred networks, a Japanese venture company, in partnership with IBM, Intel, Microsoft and NVIDIA.

Chainer 因其早期采用的“ 按运行定义 ”方案以及在大型系统上的性能而著称。第一个版本于 2015 年 6 月发布,此后在日本广受欢迎。此外,在 2017 年,它被 KDnuggets 列在十大开源机器学习 Python 项目中。

Chainer is known for its early “by run” approach and performance on large systems. The first version was released in june2015 and has since been popular in Japan. In addition, in 2017, it was listed by KDnuggets in the top ten open source machine learning Python projects.

2019 年 12 月,Preferred Networks 宣布将其开发工作从 Chainer 过渡到 PyTorch,它将仅在发布 v7 之后提供维护补丁。

In december2019, preferred networks announced the transition from chainer to pytorch, which will only provide maintenance patches after V7 is released.

按运行定义

By run definition

Chainer 是第一个引入按运行定义方法的深度学习框架。训练网络的传统过程分为两个阶段:定义网络中数学运算(例如矩阵乘法和非线性激活)之间的固定连接,然后运行实际的训练计算。这称为定义并运行或静态图形方法。Theano 和 TensorFlow 是采用这种方法的著名框架。相反,在按运行定义或动态图方法中,当训练开始时,网络中的连接是不确定的。该网络是在训练期间根据实际计算确定的。

Chainer is the first in-depth learning framework to introduce a run-by-run definition approach. The traditional process of training network is divided into two stages: defining the fixed connection between mathematical operations (such as matrix multiplication and nonlinear activation) in the network, and then running the actual training calculation. This is called defining and running or static drawing methods. Theono and tensorflow are famous frameworks for this approach. Instead, in run-by-run definitions or dynamic graph methods, the connections in the network are uncertain when the training begins. The network is determined according to the actual calculation during the training period.

这种方法的优点之一是直观且灵活。如果网络具有复杂的控制流(例如条件和循环),则在定义和运行方法中,需要针对此类构造进行专门设计的操作。另一方面,在运行定义方法中,可以使用编程语言的本机结构(例如 if 语句和 for 循环)来描述这种流程。这种灵活性对于实现递归神经网络特别有用。

One of the advantages of this method is intuitive and flexible. If the network has complex control flows (such as conditions and loops), in the definition and operation methods, special design operations are required for such constructs. On the other hand, in the run definition method, this process can be described using native structures of the programming language, such as if statements and for loops. This flexibility is particularly useful for recursive neural networks.

另一个优点是易于调试。在“定义并运行”方法中,如果训练计算中发生错误(例如数字错误),则通常很难检查故障,因为编写的代码定义了网络和实际位置。错误是分开的。在按运行定义方法中,您可以仅使用语言的内置调试器暂停计算,然后检查在网络代码上流动的数据。

Another advantage is that it is easy to debug. In the define and run method, it is often difficult to check for faults if errors occur in training calculations (such as digital errors), because the code written defines the network and the actual location. The mistakes are separate. In the run by run definition method, you can pause calculations only using the language’s built-in debugger, and then check the data flowing on the network code.

自从 Chainer 引入以来,按运行定义已经流行起来,并且现在已在许多其他框架中实现,包括 PyTorch 和 TensorFlow。

Since the introduction of chain, run by run definitions have become popular and are now implemented in many other frameworks, including pytorch and tensorflow.

扩展库

expanded memory bank

Chainer 具有四个扩展库,ChainerMN,ChainerRL,ChainerCV 和 ChainerUI。ChainerMN 使 Chainer 可以在多个 GPU 上使用,其性能明显优于其他深度学习框架。在 1024 个 GPU 上运行 Chainer 的超级计算机在 15 分钟内处理了 ResNet-50 网络上的 90 个 ImageNet 数据集,比 Facebook 以前的记录快了四倍。ChainerRL 添加了最先进的深度强化学习算法,而 ChainerUI 是一种管理和可视化工具。

Chain has four extension libraries, chainermn, chainerrl, chainercv and chainerui. The chailermn makes the chain available to use on multiple GPUs, and its performance is obviously better than other deep learning frameworks. The supercomputer running chain on 1024 GPUs processed 90 Imagenet datasets on the resnet-50 network in 15 minutes, four times faster than Facebook‘s previous records. The latest in-depth reinforcement learning algorithm is added to the chainerrl, which is a management and visualization tool.

应用程序

application program

Chainer 用作 PaintsChainer 的框架,该服务可以在用户输入最少的情况下自动对黑白(仅线条)草图进行着色。

Chainer is used as the framework for paintschainer, which automatically colors black and white (line only) sketches with minimal user input.


速搜资源网 , 版权所有丨如未注明 , 均为原创丨转载请注明原文链接:【速搜问答】Chainer是什么
喜欢 (0)
[361009623@qq.com]
分享 (0)
发表我的评论
取消评论
表情 贴图 加粗 删除线 居中 斜体 签到

Hi,您需要填写昵称和邮箱!

  • 昵称 (必填)
  • 邮箱 (必填)
  • 网址