Machine learning is now hot. As long as machine learning is applied, data and knowledge can be effectively enriched, and valuable tasks can be automated, including perception, classification, and numerical prediction. And its "brother" - the machine found, can be used to discover new knowledge that illuminates and guides mankind. Let's discuss the best application scenarios for machine learning or machine discovery, and why it is important for business.
I was a machine discovery researcher many years ago. I published academic papers in Machine Learning magazine. I also participated in machine learning related conferences and made reports because machine learning and machine discovery are similar to human activities. As an (experienced) entrepreneur, I am often asked if the learning method is very important for automatically handling certain tasks. This is why I am writing this article. Let us first review some basic concepts.
An important idea in the field of artificial intelligence is that intellectual work can be seen as a heuristic search within the “problem space†and can help find solutions to problems.
Let's imagine a common mission scenario on this TV: the detective arrived at the crime scene and the body lay on the floor. A terrible detective picks up the phone book and starts calling from the first page in order to interrogate. A very bad detective would even think that it was a space invasion or fleeing to do it, and asked NASA and the local zoo to follow these clues. Their behavior is to use the wrong heuristic method.
A good detective will be good at using the right heuristics to start with existing problems. For example: What is the cause of death? Who is the last person seen by the victim? Is there an enemy? Is there a secret romance? Debt? Good detectives will also start with the above answers to search for suspects more effectively on a large scale. The great detective may even come up with more inspiring ideas.
The key point of "machine discovery" is that discovery is just like another intelligence task. Therefore, the artificial heuristic key discovery search method applied in the problem space can also be applied to machine discovery tasks.
On the other hand, the key point of "machine learning" is to give enough data and related results, as well as some concepts (such as which data features are related to the prediction results), and then the software can also realize this association after training. Classic examples include the use of historical data to learn how to classify loan applications based on credit risk, or to predict customer loss.
What is the best application for machine learning or machine discovery?
With these key points, let us consider which design (discovery or learning) is better in specific applications. For example: Introduce traffic for large parties or events. A good party organizer needs to understand the common interests among guests and try to introduce them to each other and explain their common ground so as to promote their exchanges. This is a difficult task, so the organizers are very busy. With a list of participants, can this situation be automated?
Artificial intelligence or discovery methods do things this way: Research or find out what can lead to good mutual presentation. What determines the quality of the (referral)? Is this an innovative method of presentation for core purposes? Which data sources can enhance this automatic referral (such as LinkedIn introduction or other self-introduction)?
Then, you can generate some automatic introductions. For example, three of you graduated from the same university almost at the same time; or you all served for the African Peace Organization; even the two of you are the only ones who know machine learning here.
Bad methods of inspiration may lead to: You have both been divorced more than four times (å°´å°¬); or you both come from the Midwest (the focus is obscure); or your birthday is all winter (unrelated).
We have already discussed the key points of machine learning and machine discovery, and how to implement specific applications. Then we summarize: What is the best application for machine learning or machine discovery?
Machine discovery requires the study of the task's logic and requires corresponding knowledge, including priority paths within that range, as well as making it consistent with the actual algorithm design. This facilitates innovation in the space being searched and the heuristic method used. But the greatest innovation may come from the novel and creative output based on specific input, because automation can explore the possibility of much larger space than actually considered by humans.
Let's take a look at three examples of machine discovery engines, each of which uses programmed heuristics to explore and report as much as human knowledge can read.
In the 1990s, commercialized search engines searched for many information files, using heuristic techniques (such as page numbering, prioritization based on the content of each document or title query) to give a list of citations, and each excerpt was dynamic Customized as a function of the query word.
Around 2000, the commercialization of the classification engine put hundreds of search results in a grouped form in the theme folder, using heuristic techniques (such as the language characteristics of the topics extracted, how many search results each topic covers, The topic is divided into how the effects of the non-overlapping groups are, etc.) to describe the topics that appear in the returned search results.
In 2015, the commercialization benchmarking engine found its abnormal performance in large-scale homogenous groups, using heuristic techniques (such as combing concise, reasonable attributes, and handling abnormal pattern types) to output the ability to convey benchmarking insights about target entities. English passages.
The method of machine discovery may be: The output of the task is not just a classification or a numerical prediction. People wrote many books or articles on this task to teach new people. There is also no rich data on the input/correct output groups, so it is often convincing others why the input data and task metacognition knowledge fit the specific output. Meta-cognitive knowledge of tasks is isolated, so common sense is not needed when performing tasks.
What does this mean for technology business? Machine learning enables semi-automated automation tasks to reduce costs. Machine learning can be applied to many data-rich tasks. Machine discovery places more emphasis on specific tasks requiring specific knowledge and training. Machine discovery tends to be handmade, more elaborate and rare.
You need a lot of internal or vendor artificial intelligence expertise. Suppliers will be fewer, and they are more concerned with the specific knowledge tasks that have far-reaching impact, thus ensuring that the business is economically viable. Suppliers will not call themselves machine discovery companies. Unlike machine learning, because the machine discovers fewer companies, it is more likely to have market differentiation.
Although machine learning and machine discovery are brothers, they are separated when they are all mature.
Shenzhen ChengRong Technology Co.,Ltd. , https://www.laptopstandsuppliers.com