Programmer Think

1. Data set acquisition Data from: AI challenger 2017 image description dataset Baidu online disk: https://pan.baidu.com/s/1g1XaPKzNvOurH9M44p1qrw Extraction code: bag3 Since the original training set is too large, only the verification set AI is used here_ challenger_ caption_ validation_ 20170910.zip, unzip it 2. Text data processing ...

Note: the bugs in this article are all encountered during the installation process. After finding out the reasons, the command is optimized. Therefore, if you type according to the command, you should not encounter the following bugs It's not easy for new bloggers to sort out. If your problem is solved, please give a praise~~~~~~ I Installing ...

Introduction to machine learning for programmers (IV) - common skills in the training process This article will focus on some common skills in machine learning training using pytorch. Mastering them can make you get twice the result with half the effort. Most of the codes used will be based on the last example in the previous article, that is ...

First of all, we know that the parameters in the deep learning network are updated in the back propagation by calculating the gradient, so as to obtain an excellent parameter. However, sometimes, we want to fix the parameters of some layers and do not participate in the back propagation. For example, when fine tuning, we want to fix the paramet ...

introduction Based on PyTorch framework, this paper uses CNN convolutional neural network to realize MNIST handwritten numeral recognition, which runs only on CPU. Four structures, Linear pure Linear layer, CNN convolutional neural network, Inception network and Residual residual network, have been used to recognize handwritten digits fro ...

1, Calculation diagram 1. Forward 2. Reverse 3. Complexity The computational complexity of both forward and reverse is O (n). Because you need to traverse the graph to calculate the gradient. The memory complexity in the forward direction is O (1), while that in the reverse direction is O (n), because the reverse direction requires one ...

1. Compression of neural network For some large-scale neural networks, their network structure is very complex (it is said that some neural networks of Huawei are composed of hundreds of millions of neurons). It is difficult for us to put this neural network on a small device (such as our apple watch). This requires us to be able to compress t ...

Distributed operation Barrier of PyTorchOriginal document: https://www.yuque.com/lart/ug...On the concept of barrierFor the concept of barrier, please refer to the introduction in Wiki: synchronization barrier is a synchronization method in parallel computing. For a group of processes or threads, a synchronization barrier in the program means t ...

Five implementation strategies of pytoch's spatial shift operationThis article has authorized the platform of the polar city and is the official account of the polar platform. No second reprint is allowed without permissionOriginal document (may be further updated): https://www.yuque.com/lart/ug...prefacePreviously, I read some papers that use ...

Random talk about Autonomous vehicle_ 02 I ReadMe This paper aims to build a usable and pollution-free deep learning environment for children who are equipped with GPU in the computer and have deep learning needs.The article is based on the various methods of the great gods on the Internet, and then synthesized by myself. If there are simil ...