MIT的《深度学习》精读(17)

When designing features or algorithms for learning features, our goal is usually to separate the factors of variation that explain the observed data. In this context, we use the word “factors” simply ...
阅读(534) 评论(0)

MIT的《深度学习》精读(16)

The quintessential example of a representation learning algorithm is the autoencoder. An autoencoder is the combination of an encoder function that converts the input data into a different representat...
阅读(430) 评论(0)

人工智能的软件研发管理系统

随着人工智能的到来,在软件开发管理系统里,应该引入更加强劲的人工智能(AI),这样对整个系统更加节省人力,以及更加高效率。目前,我想到在这个方面特别适合使用目前深度学习算法来解决。比如在开发的过程中,软件开发完之后,需要交给测试人员去测试,测试人员进行测试,然后把发现的BUG上交,在这一步里,上交BUG时,其实包含了截图,这样在BUG管理平台的后台,就可以采用深度学习的方式来判断这些BUG的图片是...
阅读(541) 评论(0)

MIT的《深度学习》精读(15)

One solution to this problem is to use machine learning to discover not only the mapping from representation to output but also the representation itself. This approach is known as representation lear...
阅读(767) 评论(0)

MIT的《深度学习》精读(14)

However, for many tasks, it is difficult to know what features should be extracted. For example, suppose that we would like to write a program to detect cars in photographs. We know that cars have whe...
阅读(450) 评论(0)

MIT的《深度学习》精读(13)

Many artificial intelligence tasks can be solved by designing the right set of features to extract for that task, then providing these features to a simple machine learning algorithm. For example, a u...
阅读(371) 评论(0)

MIT的《深度学习》精读(12)

This dependence on representations is a general phenomenon that appears throughout computer science and even daily life. In computer science, operations such as searching a collection of data can proc...
阅读(598) 评论(0)

神经网络的双曲线正切激活函数

在数学中,双曲函数类似于常见的(也叫圆函数的)三角函数。基本双曲函数是双曲正弦“sinh”,双曲余弦“cosh”,从它们导出双曲正切“tanh”等。也类似于三角函数的推导。反函数是反双曲正弦“arsinh”(也叫做“arcsinh”或“asinh”)依此类推。y=tanh x,定义域:R,值域:(-1,1),奇函数,函数图像为过原点并且穿越Ⅰ、Ⅲ象限的严格单调递增曲线,其图像被限制在两水平渐近线y...
阅读(1994) 评论(0)

MIT的《深度学习》精读(11)

The performance of these simple machine learning algorithms depends heavily on the representation of the data they are given. For example, when logistic regression is used to recommend cesarean delive...
阅读(718) 评论(0)

sigmoid函数求导与自然指数

在神经网络里经常使用sigmoid做激活函数,它的导数是怎么样求解呢?因为要使用它的导数来计算梯度下降。这个过程如下:1. sigmoid函数:f(z) = 1 / (1 + exp( − z))导数:f(z)' = f(z)(1 − f(z))求导过程如下:下文解释e的来由:https://betterexplained.com/articles/an-intuitive-guide-to-ex...
阅读(7148) 评论(1)

神经元与人工神经元

生物的神经元结构如下:与人工神经元:这种现象叫做仿生学。可以使用代码来实现两个权值的感知机:w = [0, 0] b = 0 def createDataSet(): """ create dataset for test """ return [[(3, 3), 1], [(4, 3), 1], [(1, 1), -1]] def update(item):...
阅读(638) 评论(0)

MIT的《深度学习》精读(10)

The difficulties faced by systems relying on hard-coded knowledge suggest that AI systems need the ability to acquire their own knowledge, by extracting patterns from raw data. This capability is know...
阅读(408) 评论(0)

Python的sys.path妙用

在Python里,经常遇到这样一种情况,自己编写的一些模块,想比较清晰地管理不同的模块,所以在当前目录下面建立了不同的子目录,如下图:在子目录mod里包含着要使用的py文件,如果想要使用它,可以按模块的方式来导入,但是有一种更简单的方法,就是让python搜索到mod目录,即可以使用了。比如要使用上面的文件test2.py,那么就可以按下面的代码来编写:#python 3.5.3 #2017-...
阅读(383) 评论(0)

MIT的《深度学习》精读(9)

Several artificial intelligence projects have sought to hard-code knowledge about the world in formal languages. A computer can reason about statements in these formal languages automatically using lo...
阅读(461) 评论(0)

MIT的《深度学习》精读(8)

Ironically, abstract and formal tasks that are among the most difficult mental undertakings for a human being are among the easiest for a computer. Computers have long been able to defeat even the bes...
阅读(334) 评论(0)
81条 共6页1 2 3 4 5 ... 下一页 尾页
    个人资料
    • 访问:5574846次
    • 积分:72912
    • 等级:
    • 排名:第25名
    • 原创:1584篇
    • 转载:61篇
    • 译文:11篇
    • 评论:2082条
    文章存档
    最新评论