This article mainly talks about the interweaving of communication and AI, the advantages of AI in the network, and the application scenarios of AI in communication.

1. The interweaving of basic issues in communication theory and AI artificial intelligence

Basic issues in communication: accurately or approximately reproducing a message selected at another point at a certain point. Or in other words, a message is reliably transmitted from a source sender to a destination receiver by using various techniques.

AI artificial intelligence gives computers intelligence, aiming to teach them how to work, react and learn like humans. Deep learning enables human processes to absorb knowledge from data and make decisions without explicit mathematical modeling and analysis.

In order to achieve better performance in theory and practice, a major feature of communication theory is hierarchical optimization. The transmitter and receiver are usually divided into several processing modules, each processing module is responsible for specific subtasks, such as source coding, channel coding, modulation and equalization. Although this implementation is known to be suboptimal, it has the advantage that each module can be analyzed and optimized individually, resulting in a very efficient and stable system today. The current development of AI artificial intelligence can solve many non-convex optimization problems. The hierarchical optimization of communication theory may be considered as a whole, and deep learning and other means can be used to optimize our communication system as a whole, thereby achieving better performance.

2. Advantages of AI technologies such as deep learning in wireless networks

1. Processing of semi-labeled/unlabeled data

Deep learning can process large amounts of Data, and mobile networks happen to generate large amounts of different types of data quickly. Traditional supervised learning only works when enough labeled data is available. However, most current mobile systems generate unlabeled or semi-labeled data. Deep learning provides a variety of methods that allow learning useful patterns in an unsupervised manner using unlabeled data, for example, restricted Boltzmann machines (RBM), generative adversarial networks (GAN), etc.

2. Processing of geometric data

Deep learning is very effective in processing geometric movement data, which is a problem for other machine learning methods. Geometric data refers to multivariate data represented by coordinates, topological networks, measures and orders.

3. Application scenarios of ai artificial intelligence in wireless networks

1. Deep learning in fog computing

Fog computing is a concept of cloud computing As an extension, fog computing does not have the weaknesses of cloud computing introduced above. In addition, it mainly uses devices in edge networks, and data transmission has extremely low latency. Fog computing has vast geographical distribution and large-scale sensor networks with a large number of network nodes. Fog computing has good mobility. Mobile phones and other mobile devices can communicate directly with each other. The signal does not need to go around the cloud or even the base station, supporting high mobility.

2. Reinforcement learning corresponds to many communication scenarios

Many mobile network problems can be expressed as Markov decision processes (MDP), in which reinforcement learning can play an important role. Therefore, the use of deep reinforcement learning is expected to solve network management and control problems in complex, changeable and heterogeneous mobile environments.

3. Routing in communication links

Deep learning can also improve the efficiency of routing rules. Given the detailed information of routing nodes, deep neural networks are used to classify the nodes. Use deep learning technology to decide the next routing node and build software-defined routes. This significantly reduces overhead and latency, enabling higher throughput.

4. Cross-layer scheduling in communication links

For example, deep reinforcement learning is used to schedule in roadside communication networks. The interaction between the vehicle and the environment (including actions, status information, and reward signals) is formulated as a Markov decision process, and low-complexity optimization is performed by approximating the Q-value function. Compared with traditional scheduling methods, the new scheduling strategy can achieve lower interaction delays.

5. Wireless resource allocation

For example, deep reinforcement learning is used to determine the allocation of resources such as spectrum and power based on the current mode and user needs.

6. Physical Layer Security

Modern network security systems are increasingly benefiting from deep learning, as it enables systems to automatically learn signatures, patterns, and generalizations from experience To identify intrusion information (supervised learning). Compared with traditional methods, the workload can be greatly reduced and the accuracy can be improved.

To sum up, this is what the editor has compiled and shared with you today about I hope it can help everyone with the relevant content of the AI ??artificial intelligence major.