📌  相关文章
📜  PyTorch深度学习

📅  最后修改于: 2020-11-11 00:45:24             🧑  作者: Mango

深度学习

深度学习是机器学习中使用的一组算法。它是基于人工神经网络的机器学习方法的一部分。学习可以是有监督的，无监督的或半监督的。

深度学习架构(即深度神经网络，递归神经网络和卷积神经网络)已应用于自然语言处理，计算机视觉，语音识别，音频识别，社交网络过滤，机器翻译，药物设计，生物信息学，医学图像分析，材料检查和棋盘游戏程序，在某些情况下它们产生的结果要优于人类专家，甚至可以与人类专家媲美。

深度学习用于

  • 自动驾驶汽车
  • 卫生保健中的深度学习
  • 语音搜索和语音激活助手
  • 自动向无声电影添加声音
  • 自动机器翻译
  • 自动文本生成
  • 自动手写生成
  • 影像辨识
  • 自动图像字幕生成
  • 自动着色

神经网络和深度学习神经网络

人工神经网络或神经网络是根据人脑建模的。人有思维去思考并在特定情况下执行任务，但是机器如何做到这一点呢?为此，设计了一种人工大脑，称为神经网络。由于人脑具有用于传递信息的神经元，因此类似地，神经网络也具有执行该任务的节点。节点是数学函数。

神经网络基于生物神经网络的结构和功能。神经网络本身会根据输入和输出进行更改或学习。流经网络的信息由于其学习和变化特性而影响人工神经网络的结构。

深度学习神经网络是神经网络的高级形式。与简单的神经网络不同，深度学习神经网络具有多个隐藏层。深度学习神经网络会获取更复杂的数据集，因为您的模型可以学习。深度学习神经网络是

神经网络的优势

S.No Advantages Description
1. Storing information in the entire network. In traditional programming, information is stored on the entire network, not on a database. If some piece of information is missed in one place, it does not prevent the network from functioning.
2. Work with incomplete knowledge When our ANN is trained. The data may produce output either there is complete information or incomplete information. Here, the loss performance depends on the importance of the missing information.
3. Distributed memory To train an ANN, it is necessary to determine the examples, and by showing these examples train it according to the desired output. The network can produce false output is the event cannot be shown to the network.
4. Ability to make ML (Machine Learning) ANN has the capability to make a machine learn. ANN learn events and make decisions by commenting on similar events.
5. Fault tolerance features If there is a corruption in one or more cell does not prevent it from generating output, and this feature makes it fault tolerance.
6. Parallel Processing ANN can perform more than one job at the same time because of its numeric strength quality.

神经网络的缺点

S.No Disadvantages Description
1. Hardware Dependence Ann requires processors with parallel processing power according to their structure. The realization of equipment is dependent because of this reason.
2. Network’s unexplained behavior It is one of the most important problems of the ANN. It does not give any clue as to why and how, when it produces a probing solution.
3. Determination of the proper network structure For determining the structure of the neural network, there are no specific rules available. With the help of experience, trial, and error, anappropriate network structure is achieved.
4. Difficulty in showing the problem to the network ANN works with numerical information so that the problems are translated into numeric values before being introduced to ANN. For this reason, it is difficult to show the problem to the network.