Massive data brings opportunities to neural network models

Neural network model makes data more potential

We all know that in the age of mass data, deep learning offers new opportunities for artificial intelligence. These opportunities are concentrated in three places: text, pictures, and speech recognition.

Wunda mentioned that artificial intelligence has a positive cyclic chain. With a good product, you can attract more users, and then generate a huge amount of data, and then the data can bring more excellent products. However, if the traditional artificial intelligence algorithm, the data growth to a certain amount of volume, the effect of the algorithm will encounter bottlenecks. But if the new depth learning algorithm is used, the effect of the algorithm can be improved continuously with the increase of the data.

How will the neural network model be optimized? Here's an example:

7 years ago, I asked my students to use the best algorithm at the time to do a simple task-to judge which is a cup in a picture of a lot of cutlery. Basically, columnar tools are recognized as cups.

Although in the eyes of people, see the appearance of these objects, but in the eyes of the computer, they get the information is only a picture of each pixel on behalf of the color of the number, using the digital spectrum and other images contrast. To give a speech recognition example, we used to decompose the sound into different tones, phonemes, and so on, hoping to recognize a conversation through data decoding, but the result is different from the natural voice itself.

How does the human brain learn? It is stated here that, although called the neural network model, it is not true that the algorithm mimics human brain work, because we are not sure how the brain works, but only want to approach the mechanism. Take the cup above as an example, we just provide a large number of glass pictures, so that the computer to find out what the characteristics of these samples, and then it can be judged, and we usually learn in a similar way.

This thing can not be done in the past, so we need to rely on the traditional data decoding algorithm, but now the data storage and calculation are up to a huge scale, can achieve a large number of sample control.

Speech recognition will drive the mobile phone revolution

More and more people are using speech recognition, and the accuracy rate of depth learning speech recognition system is much higher than the traditional way.

In the mobile-end application scenario, the voice interaction is a more natural man-machine interaction method than the finger input. We may no longer need to install so many apps on the phone, you just have to interact with the phone, tell it what you need, and he can connect you to the service. I believe that voice will push the mobile phone revolution. Imagine, in the future, that we're going to redesign mobile products around the voice interface and redefine the interface between people and phones.

Here, Wunda said an ambiguous word: Next, we will bring you a better smartphone. Do not know whether to understand Baidu mobile phone back to the lake, or simply refer to all the mobile phone business?

Don't worry, robots will take over the world.

The artificial intelligence technology still has the challenge. For example, some countries use AI technology to threaten human rights (brain-offender tracking), which is a serious ethical issue. But even if there is no AI, this is the case. As far as the AI is concerned, it may even be in control of our world. I do not think that this is a matter of immediate concern. The machine rules the world won't happen very soon, perhaps 100 years later we need to worry about this problem.

Even if we say that the depth learning algorithm has the ability of self-learning, it does not need to worry about it. A lot of neural network algorithm model training, that is, machine learning process, are supervised training. This kind of training method is used in speech recognition, image recognition, the effect is very good. But it's a completely different thing if you stand at the point where the machine may dominate.