Artificial intelligence

 If you wish to have all the perks of being certified with the exam, you should checkout the Microsoft MS 300 Dumps offered in the ITCertDumps’s Bootcamp Program.

Artificial intelligence is now very hot. However, when everyone pays attention to "intelligence", she rarely puts the attention in "artificial", and seems to feel the artificial intelligence, everything can be automated. In fact, there is a lot of "artificial" behind artificial intelligence, and it is not automated.

1. Data Preparation

Modern artificial intelligence technology, or machine learning, its basic method and N-year data mining have nothing too much, or will A large amount of data is fed to the computer for training models. After the model is generated, it can be used for automation processing, which looks like intelligence.

However, machine learning projects for actual services are not like Alphago to generate data to train (in fact, the previous version of Alphago has used a large number of existing chess pieces), and must use the actual occurrence Data can be trained model. The models of different data trains are completely different, and the quality of the data seriously affects the effect of the model.

However, the actual data is five flowers, scattered in each application system. It is not an easy thing to organize them. Machine learning needs are often comparable to the system, which also needs to splicing the associated data in each application system; and the data coding rules of each system may be different, this also needs to be unified; some data is still original The text (log) form, it also needs to extract structured information from it in advance; not to say data from the Internet; ....

Experienced programmer knows that in a data mining project, the time for data preparation will account for approximately 70% -80%, that is, most workloads are spent in training. Before the model.

This is actually the ETL we often say, and these things seem to have a technical content. It seems that it is a programmer to do, people are not very concerned, but the cost is high.

2. Data Scientist

ETL to organize good data, is still not so easy. Data scientists are also required to further process to enter the modeling link. For example, some data is missing, then there is a way to make up; the data is too large, and many statistical methods must assume that the data distribution should try to satisfy the normal distribution, which requires the need to do it first; Business conditions generate derivative variables (such as from date to build weeks, holidays, etc.); .... Although these jobs are also prepared before modeling, they need more professional statistical knowledge, we generally do not count it as an ETL range.

The modeling algorithm for machine learning has several dozens of algorithms have their own scope, and a large number of parameters need to be adjusted. If you use a wrong model or the wrong parameters, it will get a very uncomfortable result. At this time, data scientists need data scientists to constantly try, calculate and examine data characteristics, select reasonable models and parameters, according to the results, repeated iteration, often with long time to build a practical model, short, two or three weeks, long Then March.

However, some automated modeling techniques have begun to gradually reduce the role of some data scientists, and this part is being intelligent.

We would be discussing the ways for clearing. I would suggest you focus on the below-mentioned resources and also check out the Microsoft AZ 500 Dumps offered at the ITCertDumps, they are the best when it comes to Certifications Vendor.