Four steps from acquiring AI to "solve it yourself" self -restored network

This article is by Ericsson's CIO, Diomedes Kastanis.DIOMEDES Kastanis is vp, chifeef innovation office, at ericsson, supporting advancement of the company's Technology Vision Vision and Innovation.

The expectations of machine learning (ML) and artificial intelligence (AI) are surprising.You can talk to your smartphone, and your favorite apps can identify friends and family in photos.But this is not realized overnight.The expansion of the network itself and advanced computer science and technology led them.

The most profitable fields in the current "very connected" world are information, communication, commerce, and entertainment.There are many things about the "networked society", but in the process of constantly changing, the network itself is the essential contribution of the society.

AI and ML grow the network through the current learning network to data recognition networks, self -drive networks, and self -restoration networks.

Today's network is on stage I.Solve problems with network status monitoring and optimized feedback loops close to real -time, and improve performance.The method of optimizing the perception system and network is based on the simple descriptive analysis rules and rule of thumb created by humans.For example, if signal A exceeds B of C seconds, start operating X.

These rules are usually easy to interpret, but they are hard codes (fixed), cannot adapt to changing environments, and have no complexity to effectively deal with various situations.It is not perfect as a data -driven alternative.In fact, these rules will not be found in experienced and intelligent people in a huge dataset that have meaningful correlation or impact on network KPI.In addition, "real -time" is limited and the best performance that cannot be dealt with with the optimal performance is not fulfilled by human hands.

The timing is all.Stage II networks use them to continue to find patterns in past network data and predict future actions.ML can instruct ML to analyze factors that seem to have an impact, such as time / day, network event, one or regular external event (election, natural disaster, tendency of YouTube, etc.).

The value of the data is a stochastic correlation between past network performance and manual solutions that provide future optimization.ML can obtain a correlation as if the complexity of the model allows by working with data scientists and domain experts to separate the signal from noise and test before using the ML model.The ML model can clarify the thorough distribution of the network KPI and the spacious sequence of external impact factors, and clarify the correlation to predict future results.

These predictions provide advanced advice on how to distribute network resources and execute optimization to human surveys, and improve their performance at low cost.For example, a network -type "autopilot" detects a slight prediction deviation from the optimal route and warns a human operator long before the actual problem appears.Continuous collecting data and comparing reality and predictions will improve accuracy, leading to improvement of next -generation models.

Stage II's ML method includes linear and non -linear monitoring methods, tree -based ensembles, neural networks, and batch learning (eg, pickled in overnight).Stage II predictive support is nothing but a time for human operators to change the change.The result is a breakthrough of network performance, the machine predicts, and humans can find solutions with plenty of time.

The student will eventually become a leader.

In stage III, the AI algorithm reviews past performances and identifies unrelated correlation, which will affect future performance outside human logic guidance, regardless of human instructions.Over the network data and early guidance, look at external datasets such as generated data and simulated data.

The machine uses the knowledge gained from the supervised method, applies that knowledge to a method that has not been supervised, and clarifies undetained correlation coefficients without human intervention or guidance.

The Stage III network creates the prediction of multiple patterns for administrators to determine the effects of future business based on the theoretical behavior of the network.

For example, a network gives an option to human administrators from the potential future results (optimal performance in the super bowl, or the minimum power consumption during vacation)

In this way, the network not only predicts a single future, but also opens up strategic network optimization by providing multiple foresight information to human colleagues.Stage III ML methods include deep learning technology, simulation technology, and other advanced computer science (advanced statistics, model governance, automatic model selection, etc.).

Very competent, but the Network of Stage III is not technically "intelligent."A big leap occurs in stage IV.

"I have a reason. Therefore, I am."

Stage IV networks independently identify the factors of interest that affect network performance and prioritize them.By distinguishing between predictions and (3) the most important (3) correlation relationship with the correlation, it promotes a better decision.

The distinction between the causal relationship and the correlation is based on the probable analysis that itself itself is seen in the research.AI's ability to establish a causal relationship is not the correlation, but the ability to understand the root cause of network performance.The ability to identify the causal factors will bring more accurate predictions and even better networks.At this stage, it gains the ability for the network for the cause and reason, creating a truly intelligent network.

Stage IV networks can select spontaneous actions to maximize business efficiency by directly facing the external impact.The security of the new threat can be improved, and the most commonly used KPIs can be used.This system is adapted to change in real time and continuously learns and improves with data -driven context.Stage IV ML method includes deep learning technology, enhanced learning, online learning, dynamic systems, and other advanced computer science and technology.

We propose the idea that AI and ML are partially applied before the whole is applied.There is no doubt that the world will receive great profits by the democratization and mobilization of the mountains that are constantly expanding, but it is the fastest and fastest trip to the truly intelligent machine.It is a network and a networked society.

[Original text]

[Via VentureBeat] @Venturebeat