Artificial Intelligence Framework War: FB continues to challenge Google
Artificial intelligence, how can this river and lake be calm?
Pytorch, a new open source deep learning framework, was officially released yesterday. There was a lot of hot talk on Twitter, and all kinds of artificial intelligence experts did not hesitate to say beautiful words: elegant, concise, super powerful, cool... but these are all virtual words. Why did this new Pytorch win praise?
Mainly still the "dynamic" feature.
Recently, a number of "dynamic" deep learning libraries have emerged, such as Chainer, MinPy, DyNet and so on. The benefits of dynamic libraries are obvious, and that is easy to debug and deduct. Matroid founder Reza Zadeh concluded that using the deep learning framework of Pytorch, I want to add a layer to the neural network, and I don't have to reinvent it all.
Someone gave an example on reddit. He used to build RNN (Circular Neural Network) with TensorFlow or Keras framework, but this has to specify the length of the sentence. Each instance must have the same length. Short sentences must be filled. . But if you use a dynamic library, you can handle sentences of almost any length.
The neural module network is ideal for dynamic architectures such as Pytorch. For example, specify a picture and let the artificial intelligence answer: What color is the object on the right side of the kitten in the picture? To solve this problem, including several sub-task modules: first identify the kitten, then find the object on the right side of the kitten, and then identify the specific color.
Of course, Pytorch's new features don't stop there, but it's enough to make AI engineers excited, especially Google's deep learning framework TensorFlow, which doesn't have dynamic features. Moreover, although Pytorch did not make a loud announcement, everyone knows that this deep learning framework belongs to Facebook.
Pytorch, can Facebook help Google challenge Google's dominance in the deep learning framework?
In response to the above question, Dr. Liu Shengping, senior AI technologist and head of NLP, said that the dynamic neural network is a good feature, but at present, Pytorch lacks some key features in the Beta stage, and it is difficult to shake TensorFlow in the short term. status. But he also said that the competition in the framework will continue.
Continuing to discuss this topic, I want to explain again, what is the deep learning framework. The artificial intelligence and deep learning we talked about are systems that rely on software. There are at least two ways to build this system: one is to write the code of the whole neural network step by step from scratch; in addition, it is directly assembled by using the molded module.
The so-called framework is a set of software packages that have been assembled with basic components. You can think of a deep learning framework as a set of blocks, and engineers can use different building blocks to quickly build different buildings. The differences between the various frameworks can be simply understood as the shape of the building blocks provided to the user.
At present, the mainstream deep learning framework is mostly behind different technology giants. The correspondence is as follows:
TensorFlow - Google
MXNet - Amazon
Paddle - Baidu
CNTK - Microsoft
Torch, Caffe - Facebook
In order to take the initiative in the battle for artificial intelligence, these technology giants can only compete fiercely in the field of deep learning framework. And the significance of this competition is also obvious. Whose deep learning framework can win more users, who can build a better ecosystem, and then get more vitality and faster development.
Imagine the meaning of Android to Google.
Frankly speaking, in the field of artificial intelligence, TensorFlow has obvious advantages in the deep learning framework. However, various technology giants have already recruited a large number of artificial intelligence elites. Who said that variables will not happen? Must have hope.
Two days ago, Lu Qi, who had just served as Baidu COO, talked about the artificial intelligence strategy. He also clearly stated that Baidu’s investment in artificial intelligence is not a product of AI, but hopes to build an open platform for Baidu. The technology can be applied to different fields and different enterprises.
Dr. Liu Shengping pointed out that the competition for the entrance and voice of the deep learning platform will become more and more fierce. Even if it is stronger than Google, it is unlikely to swallow a whole piece of cake in one go. For the industry, competition can avoid a single big one. .
However, it is the commercial purpose behind this technological competition to become the basis for ultimately carrying various artificial intelligence applications. Therefore, how can artificial intelligence be a calm river?
Perhaps soon, Pytorch's features will appear in TensorFlow.
Sweeping robots, also known as automatic sweepers, smart vacuum cleaners, robotic vacuum cleaners, etc., are a type of smart household appliances that can automatically complete floor cleaning in the room with a certain artificial intelligence. Generally, brushing and vacuum methods are used to first absorb the ground debris into its own garbage storage box, thereby completing the function of ground cleaning. Generally speaking, robots that complete cleaning, vacuuming, and wiping are also collectively classified as sweeping robots.
The body of the sweeping machine is a wireless machine, mainly disc type. Use rechargeable batteries to operate, the operation method is the remote control, or the operation panel on the machine. Generally, you can set the time to make an appointment for cleaning and recharge by yourself. There is a sensor in the front that can detect obstacles. If it hits a wall or other obstacles, it will turn on its own. It will take a different route according to the settings of each manufacturer and plan to clean the area. (Some earlier models may lack some functions) Because of its simple operation and convenience, it has gradually become popular nowadays, and it has become a common household appliance for office workers or modern households.
Robot technology is becoming more and more mature nowadays, so each brand has different research and development directions, with special designs, such as: double vacuum cover, attached hand-held Vacuum Cleaner, dust box can be washed and mopped, can put fragrance, or It has functions such as photocatalyst sterilization.
The fuselage is a movable device with automatic technology, and a vacuum cleaner with a dust box is matched with the fuselage to set the control path, and walk repeatedly in the room, such as: edge cleaning, centralized cleaning, random cleaning, linear cleaning, etc. , And supplemented by side brushes, central main brush rotation, rags, etc., to enhance the cleaning effect to complete the anthropomorphic home cleaning effect.
Dry Wet Robot Cleaner,Household Robot Cleaner,Portable Robot Cleaner
Shenzhen YouTai Imp.&Exp.,Co.,Ltd. , https://www.szyoutai-tech.com