AI training and inference, what are they and what do they mean?

 In the last few years, we have seen all kinds of uses for artificial intelligence , from multimedia systems that improve the resolution of images or cancel environmental noise in videoconferences, through facial recognition system and even recommendation mechanisms when we are shopping online or looking for the movie to watch at night on your Smart TV. But what is the process by which AI learns to create a model to solve these problems? This is called Training and Inference and we will explain it to you.


What can we define as Artificial Intelligence ? For some people, computers think or have their own conscience, but in the present case it has to do with the treatment of data sets, that is, with the information that is provided to the system. Today it is everywhere and it can be said that they are automated systems that generate responses from the input data they receive from a model built from a period of training or learning.

But to provide these answers, AI systems must obviously be trained, in other words: they must be educated and, therefore, taught. Let's give a little introduction to how artificial intelligence systems “learn”, what’s behind this technology, which seems advanced enough to look like magic many times.

The explanation that we are going to give you about what training and inference is is completely generic and global, so it has no specific relationship with any particular hardware.

Learn more information about the computer support specialist.

The construction of an inference model to solve a specific problem is known as training, artificial intelligence is used mainly to solve three different types of tasks: information classification tasks, pattern research tasks and automatic driving tasks. for which three different forms of learning are used:


Supervised learning: artificial intelligence receives a set of input data and the task is for AI to be able to label the input data correctly. Initially, the AI ​​receives the data set with the correct labels, and these are the training data. The generated inference model is then monitored with a set of test data, which can be answered as true or false by the AI.

Unsupervised learning: this is used when we want the AI ​​to not classify the data and therefore the labels associated with the learning data set are not used. What is sought with this learning method is to detect patterns, and what AI does in this case is to search and group the data according to their similarity. It is the most used type of learning in the processing of multimedia data.

Enhanced learning: in the specific case of this type of learning, what we do is not telling AI, for example, what a cat is or learning to distinguish a cat, but what we are doing by teaching him some rules of the game . The closest comparison in real life is when we are learning to drive a car and we have a good driving school teacher, and it is precisely AI that is used to train cars with automatic steering. In this model, the training data set is provided in real time and the conclusions drawn by the AI ​​are evaluated by a supervisory agent who feeds it. Said agent can be a human, a complex database and even another AI.

Another type of artificial intelligence is generative, which is based on the generation of data at random; These data are nothing but noise and require an external element to discard and / or classify them. In this case, the generating AI does not know what it is looking for from the beginning and ends up learning only from a second AI, which can perform the evaluation by classification or by searching for patterns.

Comments

Post a Comment

Popular posts from this blog

International English Language Testing System (IELTS)

 It has existed for almost 30 years and is one of the most widespread tests around the world. It aims to test a candidate's level of communication in English using four language skills with a real context: oral interpretation (30 minutes), text interpretation (60 minutes), writing (60 minutes) and conversation (15 minutes). Developed by the British Council in partnership with the University of Cambridge, it is applied in 16 cities in Brazil. See more:  Amazon Saheli Quiz Answers The score ranges from 0 (most basic) to 9 (most advanced). The score required by universities varies widely. However, there are reports that renowned institutions require a score higher than 8. It is also important to say that IELTS is applied in two formats: academic and general training. The academic is aimed at those people who are looking for opportunities to study master's, doctorate and postdoctoral studies abroad. General training, on the other hand, is focused on those who seek professional impr
Read more

Prevention and Recovery: How can security tools help?

 Heavy rains, floods, landslides, devastating fires, blackouts that left an entire state without electricity. Unfortunately, the list of disasters that occurred in Brazil during the year 2020 is long. Disasters appear unexpectedly, generating terrible results and, often, irreversible damage. Within a company's scenario, it is not much different. Just as nature is subject to disasters and risks, so are companies. Therefore, information security works to protect data and information within the business environment, through tools and actions that act in the prevention and recovery of data in disaster situations. In this article, you will understand how security tools can help in a disaster scenario, and why it is important to have complete and continuous actions to protect your company's data. There are two reasons why people underestimate threats and risks: ignorance and minimizing results. Although we are in the information age, the significant increase in the cost of data leaks
Read more

How to increase visibility and mitigate risks in the corporate network?

 Corporate networks are changing rapidly as companies embark on Digital Transformation, expanding on the Cloud, incorporating IoT devices and increasing wireless access. All of this increases the attack surface, making the environments more complex.  Protecting this new environment and blocking more sophisticated and intelligent cyber threats has been a major challenge for IT. Therefore, it is necessary to invest in tools capable of giving quick and accurate responses to any type of attack. The problem is that, as the context and visibility are compromised by the complexity of the environment, the IT team ends up having trouble screening many alerts at the same time.  As a result, response times are delayed and teams are often unable to accurately screen all alerts, allowing potential threats to go unnoticed. To deal with this new reality, companies need robust and efficient protection solutions, such as Gigamon's ThreatINSIGHT. It is the cloud's first native high-speed network
Read more