https://mp.weixin.qq.com/s/m9vjDGTPLWanEmbXGNXJqw
By Super Nervous
Scene description: Compared with previous years, the long-awaited "Programmers Spring Festival Gala" by many developers around the world seems to be less exciting.
Keywords: Google Search, Google Assistant, Google Lens
With much anticipation, Google I/O 2019 is here. Thousands of people attended Google’s most grand event at the Shoreline Amphitheater in California at 1 am on May 8th, Beijing time.
However, the expectations may be too high. In the two-hour speech, the "transcript" that Google handed over this year seemed to be far from satisfactory.
This year’s theme has been upgraded from the previous "AI First" to "Buiding a more helpful Google for everyone". It seems that the black technology is full of keynote speeches. After careful consideration, it feels that they have drawn a big pie.
"Transcript" display
No surprises. As predicted before the conference, at this I/O conference, Google released the following:
- Pixel 3a and Pixel 3a XL for mid-range positioning, starting at $399
- Nest Hub Max, a smart speaker upgraded to a large screen, costs $299
- Launched Android Q with multiple upgrades
- A series of developments around AI, such as the update of smart assistants, smart cameras and other applications
Judging from the press conference, Google has already laid out various fields
In addition, Google has shown its humanistic care as always this time, introducing four products/projects: Live Transcribe, Live Caption, Live Relay, and Project Euphonia. The main function is to help different people through real-time transcription, subtitles, and voice. Of disabled people realize the link with the world.
Google's humanistic care: using AI to help people with disabilities
At the end of the keynote speech, the great god Jeff Dean took the stage to talk about some of the achievements of Google AI, including applications in NLP, TensorFlow, and AI for Good.
He also focused on the application of AI in medical treatment and disaster prevention, but the content of this part has been disclosed in the Google AI summary released by Jeff Dean at the beginning of the year.
Summarizing this "transcript" in one sentence is: the hardware is understated, the software is inadequate, and the software is not enough. 
Breakthrough or stand still?
In terms of AI, the products that can be picked up at this conference are just updates that seem to be "cool" but are actually "chicken ribs".
Google Search: Make search three-dimensional
Google added 3D models and AR effects to Google Search this year. In the live demonstration, search for "shark" and a realistic 3D shark model will appear. After clicking on the AR option, you can feel the movement of the shark in the real scene up close.
The 3D model of the great white shark searched by Google, "come" to the conference stage
In addition, you can search for New Balance shoes, and then see the 3D model of the shoes and match them with your own clothes.
It looks cool, but is it really practical? Compared with the product pictures and video display of a treasure, the 3D model has little practical effect except Wow's praise. What's more, there are now live broadcasts on various shopping platforms. Isn't it more intuitive than 3D models?
Google Lens: Small updates
Google Lens became an application that can "see pictures and talk" last year. It can recognize tourist landmarks based on pictures, extract text information from photos, mobile phone numbers, and perform real-time translation.
This year's update point is mainly reflected in the function, through the mobile phone camera to obtain more information and interaction.
For example, when Google Lens is pointed at the menu, it will recommend the most popular dishes; when it is pointed at the food in the magazine, it will tell you the recipe.
Google Lens identifies restaurants based on the menu and recommends the most popular dishes
In addition, Google Lens has added a voice function, which enables translation and reading in native language.
No matter how you look at it, these functions are like toys that are painless or itchy, and their practicality is doubtful. With all kinds of food apps, do I still need it?
But in any case, Google Lens has gradually turned many ideas into reality. After that, it depends on whether the user will play with these functions.
Google Assistant: The update is good, but a 404 is a pity
Let’s talk about Google’s masterpieces. Last year, Google Assistant's phone booking function surprised everyone, but this year's update can be described as "surprise only 1 minute".
The focus of this update is super fast response speed and more complex applicable scenarios.
From the live demonstration, in less than a minute, the demonstrator used voice commands to smoothly switch multiple tasks including alarming, asking the weather, sending emails, and taking pictures.
Live demonstration by a real female assistant
It is said that this technology comes from the advancement of neural network technology. Google has developed new speech recognition and language understanding models, reducing the 100GB model in cloud computing to less than 0.5 GB.
Therefore, Google Assistant powered by AI can now run locally on the user's mobile phone. It can process voice on the phone with almost zero delay, and can be used even when there is no network connection.
In addition, Google has also put this application on the web, which can realize online car rental, and complete a series of operations such as navigation and payment.
However, after looking into it, the phone ordering function released at the Google I/O conference last year was only launched on its own Pixel phones in November 2018. Until this year, it has not been fully launched for Android and iPhone users.
For domestic users, although Google Assistant already supports the Chinese version, it is still not available on the AppStore in China. You still need to register for an account in the US to download, the so-called "404" application.
So, Google’s boasting last year has not been fully realized. Has it drawn a pie for everyone this year?
Hope the vision of Google I/O is not a string of soap bubbles
Throughout the whole keynote speech, I have to say that Google is a bit out of standard this year.
Behind the seemingly cool presentation, there are actually no major breakthroughs or upgrades in AI, and the attractive products at the I/O conference last year have not yet been fully implemented.
Google is indeed in the next big game, but judging from this report card, it is just a gorgeous PPT presentation.
The conference is full of seats, I don’t know how everyone feels after attending the conference
From this year’s Google’s vision, it can be seen that this technology giant wants to promote AI technology to all aspects and create a super collection similar to "Fulian". But in fact, the actual promotion of these applications that are chanted as "black technology" can achieve some results, depending on Google's efforts this year.
Some people complained that at this year's Google I/O conference, two products were released: one that is not available in China, and the other that is not yet ready.
Perhaps what can excuse Google is that in the past year, several AI giants have been lost. After the departure of Li Feifei and Li Jia, Google AI leader John Giannandrea also quit Apple. Not long ago, Ian Goodfellow, the father of GANs, was rumored to leave.
The big names of Google AI leaving, from left are Li Feifei, John Giannandrea, Ian Goodfellow
Google has indeed encountered a huge dilemma in the brain drain. The return of Google to China has also become more confusing.
The vision for Google to benefit everyone is truly tempting. But before a beautiful script is written into reality, perhaps what Google needs most is time.
HyperNeuropedia
Regression
Regression is a supervised learning algorithm for predicting and modeling numerical continuous random variables. It defines the relationship between input and output: in the usual regression process, the input is the existing knowledge, and the output is the prediction.
The purpose of regression is to get an optimal fitting line.
Steps to return
The steps to solve a prediction problem under the regression model are:
1) Accumulate knowledge: Call the stored knowledge Training Set.
2) Learning: Learn how to predict and get the relationship between input and output. In the learning phase, there should be appropriate guidelines, Learning Algorithm.
3) Prediction: After the learning is completed, when new data (input) is accepted, the output is predicted through the correspondence obtained in the learning stage.