algorithm - 随笔分类(第25页) - papering

Generally a good method to avoid this is to randomly shuffle the data prior to each epoch of training.

摘要：http://ufldl.stanford.edu/tutorial/supervised/OptimizationStochasticGradientDescent/ 阅读全文

posted @ 2017-09-27 12:08 papering 阅读(158) 评论(0) 推荐(0)

What is the difference between iterations and epochs in Convolution neural networks?

摘要：https://stats.stackexchange.com/questions/164876/tradeoff-batch-size-vs-number-of-iterations-to-train-a-neural-network It has been observed in practic 阅读全文

posted @ 2017-09-27 11:47 papering 阅读(195) 评论(0) 推荐(0)

But what exactly do we mean by "gets closer to"?

摘要：https://rdipietro.github.io/friendly-intro-to-cross-entropy-loss/ 【将输入转化为输出：概率分布】 When we develop a model for probabilistic classification, we aim to 阅读全文

posted @ 2017-09-26 21:00 papering 阅读(219) 评论(0) 推荐(0)

information entropy as a measure of the uncertainty in a message while essentially inventing the field of information theory

摘要：https://en.wikipedia.org/wiki/Claude_Shannon In 1948, the promised memorandum appeared as "A Mathematical Theory of Communication," an article in two 阅读全文

posted @ 2017-09-26 19:48 papering 阅读(163) 评论(0) 推荐(0)

SVM vs. Softmax

摘要：http://cs231n.github.io/linear-classify/ 阅读全文

posted @ 2017-09-26 19:23 papering 阅读(158) 评论(0) 推荐(0)

Hinge Loss

摘要：The hinge loss is a convex function, so many of the usual convex optimizers used in machine learning can work with it. It is not differentiable, but h 阅读全文

posted @ 2017-09-26 15:42 papering 阅读(375) 评论(0) 推荐(0)

recognition rate generalization识别率泛化

摘要：http://www1.inf.tu-dresden.de/~ds24/lehre/ml_ws_2013/ml_11_hinge.pdf Two extremes: • Big 𝐶 → the loss is more important → better recognition rate but 阅读全文

posted @ 2017-09-26 15:35 papering 阅读(278) 评论(0) 推荐(0)

Conditional random fields

摘要：crf 阅读全文

posted @ 2017-09-26 12:17 papering 阅读(122) 评论(0) 推荐(0)

卷积核即滤波器

摘要：输入输出的中间阅读全文

posted @ 2017-09-25 22:43 papering 阅读(189) 评论(0) 推荐(0)

Full卷积 Same卷积 Vaild卷积

摘要：阅读全文

posted @ 2017-09-25 22:03 papering 阅读(410) 评论(0) 推荐(0)

position

摘要：C:\Python36\python.exe D:/pymine/clean/chained_located/chained_located_dynamic_input.py '-69,-47,,,-72,-40,-37,-96,-36,-97,-67,-67,-43,,-100,-70,-54,-62,-92,-98,,-33,-77,-17,-17,,-98,-76... 阅读全文

posted @ 2017-09-25 18:13 papering 阅读(292) 评论(0) 推荐(0)

What happens when we continue stacking deeper layers on a “plain” convolutional neural network?

摘要：http://cs231n.stanford.edu/slides/2017/cs231n_2017_lecture9.pdf The deeper model performs worse, but it’s not caused by overfitting! 阅读全文

posted @ 2017-09-25 16:12 papering 阅读(275) 评论(0) 推荐(0)

A vectorized example

摘要：http://cs231n.stanford.edu/slides/2017/cs231n_2017_lecture4.pdf 阅读全文

posted @ 2017-09-25 14:19 papering 阅读(130) 评论(0) 推荐(0)

SVM vs. Softmax

摘要：http://cs231n.github.io/linear-classify/ http://cs231n.github.io/assets/svmvssoftmax.png 阅读全文

posted @ 2017-09-25 11:28 papering 阅读(195) 评论(0) 推荐(0)

run length encoding

摘要：直观阅读全文

posted @ 2017-09-22 16:16 papering 阅读(126) 评论(0) 推荐(0)

What I learned from competing against a ConvNet on ImageNet

摘要：http://karpathy.github.io/2014/09/02/what-i-learned-from-competing-against-a-convnet-on-imagenet/ 阅读全文

posted @ 2017-09-20 17:01 papering 阅读(189) 评论(0) 推荐(0)

Image Recognition

摘要：https://www.tensorflow.org/tutorials/image_recognition 阅读全文

posted @ 2017-09-20 16:55 papering 阅读(207) 评论(0) 推荐(0)

They're much closer in spirit to how our brains work than feedforward networks.

摘要：http://neuralnetworksanddeeplearning.com/chap1.html Up to now, we've been discussing neural networks where the output from one layer is used as input 阅读全文

posted @ 2017-09-20 16:26 papering 阅读(237) 评论(0) 推荐(0)

output value . Sigmoid neurons are similar to perceptrons, but modified so that small changes in their weights and bias cause only a small change in their output.

摘要：http://neuralnetworksanddeeplearning.com/chap1.html . Sigmoid neurons are similar to perceptrons, but modified so that small changes in their weights 阅读全文

posted @ 2017-09-20 15:18 papering 阅读(287) 评论(0) 推荐(0)

它是个黑盒，无法解释的黑盒；它避免了我们的或者某种来源的先验强加于陌生人：另外一个黑盒。

摘要：@Matthew 【[抱拳]】推荐系统中，传统的两大算法，无论是基于人的过滤，还是基于物品的过滤，在前期历史数据的量和质都不充足的情况下，这两种传统的算法都无法冷启动。【基于ANN的一种解决办法】张三第一次点击了汽车广告，通过ANN训练，使得输出层的汽车的概率最大，次之的就为接下来要投出的广告类阅读全文

posted @ 2017-09-19 20:28 papering 阅读(221) 评论(0) 推荐(0)

papering

随笔分类 - algorithm