In the AI ​​boom, the voice of the neural network is the loudest. However, AI is much more than that.
At present, in the field of AI technology, the most invested is the research on neural networks. In the eyes of everyone, neural network technology seems to be "the brain of program construction" (although the metaphor is very inaccurate).
The concept of neural networks was proposed as early as the 1940s, but until now, people still know very little about the way neurons and brain work. In recent years, the research community has become more and more vocal about neural network technology innovation. Desire to restart the neural network...
In fact, in addition to the neural network, the AI ​​field also contains a lot of more interesting, newer and more promising technologies, which are introduced to you in the article.
Knol extractionKnol refers to information units, that is, keywords, words, etc. Knol extraction technology is the process of extracting key information from text. To give a simple example: For example, "As the name implies, the octopus has 8 legs", after the phrase is extracted, it becomes like this: {"fish octopus": {"number of legs": 8}}.
Our commonly used Google search engine relies on this technology, and many of the technologies introduced later include this technology.
2. Ontology constructionOntology construction is an NLP-based technology designed to build a hierarchy of entity nouns using software. This technique is very helpful in implementing AI sessions. Although the ontology build surface looks simple, the fact is that construction is not easy, mainly because the actual connection between things is more complicated than we think.
For example, use NLP to analyze text to build an entity relationship set:
Example: "My Labrador has just given birth to a group of puppies, their father is a poodle, so they are Labrador Poodles (a kind of mixed-breed dog)" Become: {"Puppy å´½": {"may be": "Labrador Poodle", "have": "Father"}, "Labrador": {" own /have": "puppy å´½"}}.
However, when humans express their language, they usually do not state all the relationships. For example, in this sentence, it is necessary to infer that "my Labrador is a female", this is The difficulty of ontology construction.
As such, ontology building technology is currently only used in top-notch chat bots.
Heuristics are rules for classification, usually similar to conditional statements such as "if the item is red" or "if Bob is at home", these conditional statements are often accompanied by an action or decision, such as:
If the ["component"] attribute contains the element "arsenic": its ["poison"] attribute is "True".
For each new information, new heuristics and new relationships are accompanied, and with the establishment of new heuristics, new understandings of related nouns can be made. such as:
Heuristic 1: "puppies" (puppies) instructions are Babies;
Heuristic 2: Babies are very young;
Inferred from the above two heuristics: "puppies" are very young.
The difficulty with heuristics is that, in most cases, the rules are not as simple as "If/Then". A statement like "some people have golden hair" is difficult to express in heuristics. So we have "cognition" (see below).
4. EpistemologyEpistemology is a combination of ontology construction and custom heuristics, and incorporates a probabilistic feature, which expresses the possibility that a noun can be associated with any attribute by probability. For example, use this ontology structure:
{'person': {'gender': {'male': 0.49, 'female': 0.51}, 'race': {'Asian': 0.6, 'African': 0.14}}
To indicate the judgment of a person's gender and race. At the same time, probabilities can help identify “hybrid†phrases with multiple meanings, such as “the plums are like hormonal raisins,†because the phrase “playing hormones†is likely to mean “ The larger the volume, the more likely it means that the plum is bigger than the raisins.
The realization of epistemology is much more difficult than the ontology construction. First, it requires more data; and, due to its structural complexity, it is difficult to quickly establish a database to determine the lookup after determining the rules; and, the determination of the rules is usually based on a thing being mentioned in a paragraph of text. And the frequency, but the text may not be able to truly reflect the reality.
Epistemology is very similar to the "tensor flow" theory proposed by Asimov. The same-named TensorFlow system developed by Google is not really based on tensor, and epistemology is based on tensor.
5. Automatic gauge technologyA gauge system must contain the appropriate evaluation criteria. Imagine that when you buy a house, there are factors such as housing size, location, price, and style. These factors are not necessarily positive, and you need to make decisions by measuring trade-offs. For example, if you care more about the housing area than the price, you would rather spend a few times more money to buy a big house.
Self-assessment techniques determine the weight of each factor by how much you value different factors to make decisions. Through this process, you can also predict inventory changes, recommend products, and achieve automatic driving. That is to say, most of the functions that neural networks can achieve, automatic gauge technology can be competent, although it takes longer training time, but it has several orders of magnitude faster decision-making speed.
6. Vector differenceVector difference techniques are commonly used for image analysis and for the processing of time-varying data. By constructing an abstract vector image for the target, the candidate object is compared with the target object to be identified, thereby determining whether it is "the best dating face" or "the best buying opportunity".
Usually, the difference between the target objects is accompanied by a quantitative rule that measures the degree of difference. Through the vectorization of features, some concepts of “fuzzy†are simply and clearly expressed.
For example, for humans, we generally think that symmetrical faces are more attractive, but for computers, accurate calculations are needed to judge, and at this time, facial abstraction is performed by 30 triangles, rather than through intact faces. Image comparisons can save a lot of computation time and storage space.
Processing of non-image data is also possible. For example, stock price changes, the ratio of earnings per share to margin, etc., by vectorizing these data and comparing it with the ideal value, you can determine the profit or risk of an investment.
7. Matrix convolutionConvolution matrices are often used for edge detection and contrast enhancement in image processing. For example, many of the filters in PhotoShop are based on convolution matrices or superimposed convolutions (multiple convolution operations in a specific order).
At the same time, the convolution matrix can also be used to process non-image data. For example, when a time series vector is processed using a convolution matrix, the pattern can be quickly found like edge detection, and then a specific value or range can be found at the minimum or maximum value to make a judgment.
8. Multi-view decision systemThe decision to make a decision is not simple. The multi-view decision system makes decisions in multiple ways in a more democratic form.
For example, in the case of a house, your optimism about a house may be based on a lack of comprehensive factors, and the subsequent "this house is built on a cliff" (of course, this overwhelming factor may come from Knol extraction will eliminate all your previous good feelings and let you make a decision again.
Therefore, decision-making needs to be considered through more comprehensive factors, and multi-view decision-making systems can use two sets of standards (such as you and your spouse) to measure decisions. The multi-view decision system can also be applied to the field of automatic driving, for example, collecting opinions of 10,000 car owners to develop new standards.
Written at the end - believe that the technology does not press
Many people have only one tool in their eyes, falling into the deep pit of "I have a hammer, so everything is a nail." Companies such as Recognant, while applying neural networks, also apply these relatively unpopular techniques in the application article, after all, compared to neural network hardware systems.
The advantage of these software technologies is that they can be adjusted and developed at any time for different situations without additional cost. Therefore, if the technology is narrow, it may be trapped by some situations, and the wider the technical surface, the easier it will be to solve the problem.
The 8-inch tablet will have a big impact on the 7-inch and 10-inch tablet market. Because the portability of an 8-inch tablet is stronger than that of a 10-inch tablet, and the usable area is larger than that of a 7-inch tablet. The most important thing is that the price is more moderate, which is much cheaper than a 10-inch tablet. It can be said that the 8-inch tablet computer has a good balance between portability and screen display area, and is more likely to be favored by the majority of users.
8 Inches Tablet Pc,Tablet Computer,8 Inch Android Tablets,8 Inch Tablet
Jingjiang Gisen Technology Co.,Ltd , https://www.jsgisengroup.com