Three categories of artificial intelligence

Transferred from: http://it.sohu.com/20161102/n472067196.shtml

Ai has come into all areas-from autonomous cars, to automatically replying to e-mails, to smart homes. You seem to be able to get any product (such as health, flight, travel, etc.) and make it smarter through the special application of artificial intelligence. So unless you believe that the event has a Terminator twist, you may ask yourself what the AI can do to herald the interests of the workplace or the overall line of business.
There are three main branches of AI:

1) Cognitive ai (cognitive ai)
Cognitive computing is the most popular branch of AI, responsible for all the interactions that feel "like human beings." Cognitive AI must be able to handle complexity and two semantics with ease, while continuing to learn from the experience of data mining, NLP (Natural language Processing), and intelligent automation.
There is a growing tendency to think that cognitive AI mixes the best decisions made by AI and the decisions of human workers to oversee more difficult or uncertain events. This can help expand the applicability of AI and generate faster, more reliable answers.

2) Learning AI (machine learning ai)
Machine learning (ML) AI is the kind of artificial intelligence that can automatically drive your Tesla on a freeway. It is also at the forefront of computer science, but it is expected to have a great impact on everyday workplaces. Machine learning is to look for "patterns" in big data and then use these patterns to predict results without too much human explanation, and these patterns are not visible in ordinary statistical analysis.
However, machine learning requires three key factors to be effective:
A) data, a large amount of data
To teach AI new tricks, a lot of data needs to be fed into the model to achieve a reliable output score. For example, Tesla has deployed auto-steering features to its car while sending all the data it collects, drivers ' interventions, successful escapes, and false alarms until headquarters, learning and sharpening the senses in error. A good way to generate a lot of input is through the sensor: whether your hardware is built-in, like a radar, camera, steering wheel, etc. (if it's a car), or you're inclined to the internet of Things (things). Bluetooth beacons, health trackers, smart home sensors, public databases, and more are just a few of the more connected sensors in the Internet, which can generate a lot of data (too much for any normal person to handle).
b) found
To understand data and overcome noise, machine learning algorithms can sort, slice, and translate confusing data into understandable insights. (If you want to scare off your coworkers, listen to the different sorting algorithms that are used in the first)
Video full length 5:50, please watch under the WiFi condition, local tyrants casual
There are two kinds of algorithms learned from data, unsupervised algorithms and supervised algorithms.
Unsupervised algorithms only process numbers and raw data, so descriptive labels and dependent variables are not established. The goal of this algorithm is to find an intrinsic structure that people do not expect. This is useful for in-depth understanding of market segments, correlations, outliers, and more.
On the other hand, supervised algorithms use these relationships to predict future data by knowing the relationships between different datasets through tags and variables. This can be useful in climate change modelling, predictive analytics, content recommendations, and more.
c) Deployment
Machine learning needs to go from computer science labs to software. More and more vendors like CRM, Marketing, ERP, etc., are improving the ability of embedded machine learning or close integration with the services that provide it.

3) deep Learning (Deepin learning)
If machine learning is cutting-edge, deep learning is cutting-edge. This is an AI you will send it to the quiz. It combines the analysis of big data and unsupervised algorithms. Its applications typically revolve around a large, unlabeled set of data that needs to be structured into interconnected clusters. This inspiration for deep learning comes entirely from the neural networks in our brains, and is therefore aptly referred to as artificial neural networks.
Deep learning is the foundation of many modern speech and image recognition methods, and has higher accuracy over time than previously available non-learning methods.
We hope that in the future, deep learning AI can answer customers ' inquiries and complete orders via chat or email. Or they can help with marketing on recommended new products and specifications based on their huge data pools. Or maybe one day they can become a full-service assistant in the workplace, completely blurring the boundaries between robots and humans.
Artificial intelligence survives and improves by the scale of data used on it, which means that not only can we see better AI over time, but they will evolve around organizations that can tap the largest data sets.